Contributions to the evaluation and improvement of LoRaWAN

PhD Thesis Contributions to the Evaluation and Improvement of LoRaWAN by Lluís Casals Ibáñez Advisors: Carles Gómez Montenegro and Rafael Vidal Ferré PhD Program of Network Engineering Network Engineering Department Universitat Politècnica de Catalunya Castelldefels, March 2023 i Abstract Low Power Wide Area Network (LPWAN) technologies have gained significant momentum as wireless solutions for implementing Internet of Things (IoT) solutions. The unique set of LPWAN features, such as long range, low power consumption, and low infrastructure cost, along with challenging message sizes and message rates, has attracted the attention of the industry, academia and standard development organizations. Within the LPWAN family, LoRaWAN has become a very popular technology. After the publication of the LoRaWAN specification and the availability of certified hardware, many researchers have devoted their efforts to investigate the performance of this technology. Since the IoT is expected to include a very large number of devices (such as sensors and actuators), and LoRaWAN is expected to support up to hundreds of thousands of IoT devices per radio gateway, the energy efficiency, performance and scalability of LoRaWAN are hot research areas. This PhD thesis focuses its research on a set of crucial topics around the aforementioned areas. Many LoRaWAN devices, such as sensors or actuators, will typically run on battery power. Therefore, it is crucial to investigate the power consumption characteristics of LoRaWAN. However, published works only focus on this topic to a limited extent, providing only rough estimates of parameters related to LoRaWAN energy performance, without considering the realistic behavior of the LoRaWAN device hardware, as well as the impact of the main LoRaWAN parameters and mechanism settings. Therefore, one of the objectives of this PhD thesis is to propose an analytical model that characterizes device current consumption, lifetime and energy cost of data delivery. On the other hand, the study of the behavior of confirmed data transmission and the evolution of the Spreading Factor (SF) parameter due to the rules defined in the LoRaWAN specification in case of retransmissions is also in our interest. Through simulations, we study the behavior of LoRaWAN under controlled conditions, for different load conditions or different active LoRaWAN functionalities (e.g. confirmed versus unconfirmed data transmission). Starting with the FLoRa simulator, based on OMNeT++, we improved and completed this simulator to include the unimplemented functionalities and correct some behaviors that did not conform to the LoRaWAN specification. In a preliminary analysis, we realize that SF tends to downgrade to SF12, corresponding to the lowest data rate and the longest data transmission time, which leads to poor network performance, although it could be, in some conditions, more reliable. We have called this phenomenon the "SF12 well". In this context, we have considered alternative mechanisms to update the SF parameter in the case of data retransmissions, which allows to improve the transmission in the uplink channel. As the third objective of the present PhD thesis, we focus on the impact of the packet size on the energy efficiency in LoRaWAN. Furthermore, to understand the observed behavior, we also consider the impact of packet size on other performance parameters. Existing studies on LoRaWAN energy efficiency evaluation, including those that consider packet size as a parameter, present significant limitations. We also Contributions to the Evaluation and Improvement of LoRaWAN ii carry out our study by means of the simulator we developed to address the above objective, with additional modifications to compute energy consumption in a more detailed process. In this work we also apply the realistic energy model obtained in the first proposed objective. As a complementary work, we also evaluate the influence of duty cycle constraints on the overall performance of LoRaWAN and specifically on the energy performance. Abstract iii Resum Les tecnologies de xarxa d'àrea extensa de baixa potència (LPWAN) han guanyat un impuls significatiu com a solucions sense fil per implementar aplicacions i serveis d'Internet de les coses (IoT). El conjunt de característiques únic de LPWAN, com ara llarg abast, baix consum d'energia i baix cost de la infraestructura, juntament amb els desafinaments que representen la mida dels missatges i les taxes de missatges, ha acaparat l'atenció de la indústria, el món acadèmic i les organitzacions de desenvolupament d’estàndards. Dins de la família LPWAN, LoRaWAN s'ha convertit en una tecnologia molt popular. Amb la publicació de l'especificació LoRaWAN i la disponibilitat de maquinari certificat, molts investigadors han dedicat els seus esforços a estudiar el rendiment d'aquesta tecnologia. Atès que es preveu que la IoT inclogui un nombre molt elevat de dispositius (com sensors i actuadors), i que s'espera que LoRaWAN admeti fins a centenars de milers de dispositius IoT per passarel·la ràdio, l'eficiència energètica, el rendiment i l'escalabilitat de LoRaWAN són àrees d'investigació candents. Aquesta tesi doctoral centre les seves investigacions en un conjunt de temes crucials al voltant d’aquestes àrees. Molts dispositius LoRaWAN, normalment, funcionen en base a una bateria. Per tant, és de vital importància investigar les característiques del consum d'energia de LoRaWAN. Tanmateix, els treballs publicats només fan estudis limitats, proporcionant només estimacions aproximades dels paràmetres relacionats amb el rendiment energètic, sense tenir en compte el comportament realista del maquinari dels dispositius LoRaWAN, o l'impacte dels principals paràmetres de LoRaWAN i la configuració del mecanisme. Per tant, un objectiu del present treball és proposar un model analític que caracteritzi el consum real dels dispositius, la seva vida útil i el cost energètic del lliurament de dades. D'altra banda, també abordem l'estudi del comportament de la transmissió de dades amb confirmació i l'evolució del paràmetre Spreading Factor (SF) segons les regles definides a l'especificació LoRaWAN en cas de retransmissió. Mitjançant simulacions, hem estudiat el comportament de LoRaWAN en condicions controlades de càrrega o de funcionalitats actives de LoRaWAN (p. e., transmissió de dades confirmada versus no confirmada). Prenent com a base el simulador FLoRa, basat en OMNeT++, hem millorat i completat aquest simulador per incloure les funcionalitats no implementades, i corregir alguns comportaments que no s'ajustaven a l'especificació de LoRaWAN. L'anàlisi dels resultats preliminars ens indica que SF tendeix a canviar cap a SF12, que correspon a la velocitat de dades més baixa i al major temps de transmissió de dades, la qual cosa comporta un baix rendiment de la xarxa, encara que podria ser, en algunes condicions, més fiable. A aquest fenomen l'hem anomenat "pou SF12". En aquest context, hem considerat mecanismes alternatius a l'hora d’actualitzar el paràmetre SF en el cas de retransmissions de dades, que permet millorar la transmissió en el canal de l'enllaç ascendent. El tercer objectiu d'aquesta tesi és estudiar l'impacte de la mida del paquet en l'eficiència energètica de LoRaWAN. A més a més, per entendre el comportament observat, també considerem l'impacte de la mida del paquet en altres paràmetres de rendiment. Els estudis existents sobre l'avaluació de l'eficiència energètica de LoRaWAN, inclosos els que consideren la mida del paquet com a paràmetre, presenten Contributions to the Evaluation and Improvement of LoRaWAN iv limitacions significatives. El nostre estudi també el realitzem mitjançant el simulador que hem desenvolupat per abordar l'objectiu anterior, amb modificacions addicionals per tal de calcular el consum d'energia d'una manera més detallat. En aquest treball també apliquem el model energètic realista obtingut en el primer objectiu plantejat, i avaluem la influència de les restriccions del cicle de treball en el rendiment general de LoRaWAN i en el rendiment energètic, en particular. v Agraïments / Acknowledgments En primer lloc, vull agrair la guia i l'ajuda dels meus directors de tesi: Carles Gómez i Rafael Vidal. Si miro enrere i observo tot el camí que hem fet, estic segur que no hauria arribat fins aquí sense la vostra expertesa, suport i, també, sense la vostra flexibilitat en els moments que ha calgut compaginar tesi, docència i família. Tot plegat ha estat un període molt intens i enriquidor, que ha estat possible per la vostra vàlua com a professionals però també com a persones. Molt sincerament, GRÀCIES! A tots els companys de departament, especialment a la secció de Castelldefels-Vilanova per la seva col·laboració en els moments que ha calgut, sobre tot en l'última etapa d'aquest treball, fent possible compatibilitzar la càrrega docent que tenia a l'escola i la dedicació a la tesi doctoral. Al grup de recerca WNG, on he desenvolupat tota la feina d'investigació relacionada amb aquesta tesi i altres activitats al llarg dels anys, gràcies per permetre'm col·laborar amb tots vosaltres, i per donarme una referència per continuar treballant, i fer-ho en un ambient positiu i constructiu. I would like to thank the external experts for reviewing this PhD thesis, as well as their comments and suggestions to improve it. In addition, I also thank the members of the PhD dissertation committee for their availability and time in reviewing and evaluating this work. Vull fer una menció a totes les eines (o les més importants) que han fet possible, o han facilitat, que pogués desenvolupar totes les tasques relacionades amb aquesta tesi: Linux, Python (amb Matplotlib, Pandas, Numpy), OMNeT++, Bash, awk, sed, C, etc. No se si aquestes eines haurien sigut capaces de fer aquesta tesi. Del que si que n'estic segur és que sense elles, jo encara estaria comptant paquets de LoRaWAN. Gràcies a totes els seus desenvolupadors que han fet possible que tingués una feina més senzilla. També vull agrair les mostres d'interès i ànims de tota la família i tots els amics que han seguit el meu camí per aquest doctorat. Això de fer una tesi doctoral, fora del nostre àmbit professional o acadèmic, sempre té una aura de misteri. Malgrat això, es d'agraït l'esforç per acostar-se a l'ocupació que m'ha tingut absorbit durant els darrers any. Per acabar, i molt especialment, vull agrair a la Carme i la Laia el vostre suport, paciència, flexibilitat i comprensió. A estat un llarg camí (molt llarg si també comptem els intents fallits) on "la tesi" ens ha acompanyat a tot arreu com una maleta plena de feina extra. Espero que el final i el resultat us hagi complagut també a vosaltres. MOLTES GRÀCIES per ser-hi. vii Contents List of Figures ......................................................................................................................... ix List of Tables ........................................................................................................................ xiii Glossary ................................................................................................................................ xv 1. Introduction ....................................................................................................................... 1 1.1. Motivation ......................................................................................................................................... 1 1.2. Objectives ........................................................................................................................................ 1 1.3. Results and contributions ................................................................................................................. 2 1.4. Organization of this PhD thesis ........................................................................................................ 3 2. LoRaWAN overview ............................................................................................................ 5 2.1. LoRaWAN general features ............................................................................................................. 5 2.2. LoRaWAN physical layer ................................................................................................................. 6 2.2.1. Receive window parameters ................................................................................................. 8 2.3. LoRaWAN MAC layer ...................................................................................................................... 9 3. Modeling the energy performance of LoRaWAN ............................................................... 11 3.1. Related work .................................................................................................................................. 11 3.2. Modeling LoRaWAN ED current consumption ............................................................................... 14 3.2.1. Unacknowledged transmission ............................................................................................ 14 3.2.2. Acknowledged transmission ................................................................................................ 20 3.3. Evaluation ...................................................................................................................................... 24 3.3.1. ED current consumption in unacknowledged transmission ................................................. 25 3.3.2. ED current consumption in acknowledged transmission ..................................................... 26 3.3.3. ED lifetime ........................................................................................................................... 29 3.3.4. Energy cost of data delivery ................................................................................................ 30 4. The SF12 Well in LoRaWAN: problem and ED solutions ..................................................... 39 4.1. Related work .................................................................................................................................. 39 4.2. The problem ................................................................................................................................... 41 4.2.1. Simulator details .................................................................................................................. 42 4.2.2. Simulated scenarios ............................................................................................................ 42 xv xv Glossary ACK_TIMEOUT Acknowledgment Timeout AFLoRa Advanced Framework for LoRa ADR Adaptive Data Rate BER Bit Error Rate BLE Bluetooth Low Energy CAD Channel Activity Detection CED Confirmed mode ED CR Coding Rate CRC Cyclic Redundancy Check CSS chirp spread spectrum DR Data Rate ETSI European Telecommunications Standards Institute ED End-device EDL End-Device Load EPB Energy consumption Per delivered data Bit EU European Union FHDR Frame Header FLoRa Framework for LoRa FPort Frame Port FRM Payload Frame Payload GFSK Gaussian Frequency Shift Keying IoT Internet of Things IP Internet Protocol ISM Industrial Scientific Medical LoRa Long Range LoRaWAN Long Range Wide Area Network LPWAN Low Power Wide Area Network MAC Medium Access Control MAX_RETR Maximum number of message retransmissions by an ED MHDR MAC Header MIC Message Integrity Code OTAA On-The-Air Activation PDR Packet Delivery Ratio PHDR Physical Header PHDR_CRC Physical Header Cyclic Redundancy Check PHY Physical QoS Quality of Service RECEIVE_DELAY1 Delay from end of uplink transmission to start of 1st receive window RECEIVE_DELAY2 Delay from end of uplink transmission to start of 2nd receive window RF Radio Frequency RX1 Receive Window 1 RX1DROffset Offset of DR in the first receive window, RX1 RX2 Receive Window 2 SF Spreading Factor SINR Signal to Interference and Noise Ratio SPI Serial Peripheral Interface ToA Time-on-Air TP Time between Packets UED Unconfirmed mode ED WFT Well Fall Time 1. Introduction This chapter provides the motivation, the contributions and the organization of this PhD thesis. We start with the motivation for this work in Section 1.1. We follow with the description of the main contributions from this PhD thesis in Section 1.2. Finally, we detail the chapters of this PhD thesis document in Section 1.3. 1.1. Motivation Low Power Wide Area Network (LPWAN) technologies have gained significant momentum as wireless solutions for enabling Internet of Things (IoT) applications [1,2]. The unique set of LPWAN features, such as long range, low energy consumption and low infrastructure cost, along with challenging message sizes and message rates, has attracted the attention of the industry, academia and standard development organizations [2,3]. Within the LPWAN family, LoRaWAN has become a very popular technology [4,5]. After the publication of the LoRaWAN specification [6] and the availability of certified hardware [7], many researchers have devoted their efforts to investigate the performance of this technology. Since the IoT is envisaged to involve a massive number of devices (such as sensors and actuators), and LoRaWAN is expected to support up to hundreds of thousands of IoT devices per radio gateway, energy efficiency, performance and scalability of LoRaWAN is a hot research area [4,5,8–27]. This PhD thesis carries out research in a set of crucial topics. We next overview the main research objectives of this PhD thesis. 1.2. Objectives Many LoRaWAN devices, such as sensors or actuators, will typically be battery-operated. Therefore, it is crucial to investigate the characteristics of LoRaWAN energy consumption. However, published works only focus on this topic to a limited extent, providing only rough estimates on parameters related with LoRaWAN energy performance, while not considering the realistic behavior of LoRaWAN device hardware, as well as the impact of the main LoRaWAN parameters and mechanism settings [14,23,26,27]. This will be one objective of the present work: to propose an analytical model that characterizes device current consumption, lifetime and energy cost of data delivery. Contributions to the Evaluation and Improvement of LoRaWAN 2 On the other hand, the study of the behavior of confirmed data transmission and the evolution of the Spreading Factor (SF) parameter due to the rules defined in the LoRaWAN specification in case of retransmissions is also in our interest. As we will detail in Chapter 4, it is advantageous to use some kind of simulator to study this behavior of LoRaWAN in controlled conditions, in order to set known initial parameters that enable to compare the same system under different load conditions or different LoRaWAN active functionalities (e.g., confirmed vs unconfirmed data transmission). We found a set of simulators used in LoRaWAN evaluation studies, which however are very focused on a partial model of LoRaWAN. Therefore, our choice has been to build a new simulator, by extending the functionality of an existing one, FLoRa [28], based on OMNeT++. FLoRa implements the most basic functionality of physical and Medium Access Control (MAC) levels of LoRaWAN, and is only able to simulate transmission in the uplink channel and a limited transmission in the downlink channel. In a preliminary analysis, we realize that SF tends to downgrade to SF12, corresponding to the lowest data rate and greatest data transmission time, that leads to a low performance of the network, although it could be, in some conditions more reliable. In this context, we have considered alternative mechanisms to update the SF parameter, which allows to improve the transmission in the uplink channel. As a third objective of our PhD, we focus on the impact of packet size on the energy efficiency in LoRaWAN. However, to understand the observed behavior, we also consider the impact of packet size on several other performance parameters. Existing studies on evaluating the energy efficiency of LoRaWAN, including those that consider packet size as a parameter, present significant limitations. We carry out this study by means of the simulator we developed for the second objective, with additional updates in order to compute the energy consumption in a more detailed process. In this work we also apply the realistic energy model obtained in the first objective of the PhD thesis research. As a complementary work, we evaluate the influence of the duty cycle restrictions on the general performance of LoRaWAN and, specifically, on the energy performance. 1.3. Results and contributions The main contributions of this PhD thesis are the following: A) Developing an energy model for LoRaWAN networks. B) Characterizing the SF12 Well problem in a LoRaWAN network, and proposing (and evaluating) a number of solutions. C) Studying the packet transmission energy efficiency in LoRaWAN and determining the optimal packet size in several conditions. D) Determining the influence of the duty cycle restriction on the energy performance and in the communication performance in LoRaWAN networks. E) Improving a network simulator for LoRaWAN including acknowledged transmission mode, and several SF management methods. 1. Introduction 3 1.4. Organization of this PhD thesis The remainder of the document is organized as follows. Chapter 2 overviews LoRaWAN, describing its general architecture and focusing on its physical and MAC layer details. In Chapter 3, we present a model for the energy performance of LoRaWAN. Chapter 4 describes and discusses the causes of the SF12 Well problem, and proposes solutions to this phenomenon. Chapter 5 is focused on the study of packet size energy efficiency-optimization, considering a wide set of network conditions. In Chapter 6, we present the conclusions and future works from this PhD thesis. 2. LoRaWAN overview In this chapter, we present fundamental LoRaWAN characteristics. We describe the protocol architecture as well as the physical and MAC layers, highlighting the mechanisms, procedures and key parameters that are relevant in the scope of this PhD thesis. This chapter is organized in three sections. The first one provides a general LoRaWAN overview, whereas the remaining two sections focus on LoRaWAN physical and link layer functionality, respectively. 2.1. LoRaWAN general features LoRaWAN is a wireless communication technology that offers long range (often in the order of kilometers) [5] while supporting low energy consumption (e.g., allowing multiyear lifetime for batteryoperated devices) [29]. As in other LPWAN technologies, long range is achieved at the expense of reduced communication capacity. However, this feature does not pose a problem for many IoT use cases. The LoRaWAN network architecture comprises three main types of network entities: end-devices (EDs), gateways, and a network server (NS) (see Figure 1a). These elements are organized in a topology known as star of stars [30]. The EDs typically correspond to constrained devices such as sensors. The EDs transmit LoRaWAN messages to the NS as their destination endpoint through one or more gateways. This type of message transmission is known as uplink transmission. In the downlink, the NS may transmit LoRaWAN messages to the EDs through only one gateway. Communication between the EDs and the gateways is carried out by means of a physical layer called LoRa. The gateways and the NS are connected by means of an IP-based network, while the gateways forward LoRaWAN messages between the EDs and the NS (see Figure 1b) [29]. The NS centralizes data collection, which allows a separate application entity (e.g., an application server) to access the NS data. LoRaWAN defines three classes in terms of supported features and functionality: class A, class B, and class C. Class A, which is also referred to as basic LoRaWAN, is mandatory for all LoRaWAN devices. In this class, and for the sake of energy saving, downlink transmission is only allowed in time intervals called receive windows, which are subsequent to an uplink transmission. Class B is defined on the basis of class A, offering additional downlink transmission opportunities at times which may be scheduled a priori. In contrast with class A and class B, class C offers no constraints for downlink transmission. However, the class C EDs cannot turn off their radio interface and are not suitable for devices with a constrained energy Contributions to the Evaluation and Improvement of LoRaWAN 6 source. Most LoRaWAN devices implement only class A features, as the other classes are optional. In this PhD thesis we assume that class A is used in all the considered LoRaWAN scenarios. (a) (b) Figure 1. LoRaWAN system. (a) General architecture. (b) Protocol stack for each network element. The next two subsections overview the main features of the class A LoRaWAN physical layer and the MAC layer, respectively. 2.2. LoRaWAN physical layer LoRaWAN supports two types of modulations for physical transmission between an ED and a gateway: the LoRa modulation and the Gaussian Frequency Shift Keying (GFSK). The former is the most frequently used modulation in many scenarios. LoRa is based on a chirp spread spectrum (CSS) [31]. The duration of a LoRa symbol depends on the SF in use, as a LoRa symbol comprises 2SF chips [5]. Six SFs PHY LoRa PHY LoRa End-Device (ED) Gateway Network server (NS) PHY PHY LoRaWAN Customer application LoRa or FSK Modulation Ethernet, Wi-Fi, 4G, etc. MAC MAC Encrypted Encrypted Backhaul IP Stack Backhaul IP Stack Customer application LoRaWAN Message forwarding 2. LoRaWAN overview 7 (ranging from seven to twelve) are defined, leading to six different corresponding Data Rates (DRs) (see Table 1). The SFs are orthogonal, which contributes to spectral efficiency. LoRa modems also use forward error correction, adding a small overhead to the transmitted message, which provides recovery features against bit corruption. This is implemented through different Coding Rates (CRs), from 4/5 to 4/8 (denoted CR = 1 to CR = 4, respectively). On the other hand, to avoid issues regarding drift of the crystal reference oscillator, a low DR optimization mechanism is applied, which adds a small overhead, to increase robustness to frequency variation over the timescale of the LoRa message [31]. This is done for SF = 11 and SF = 12. LoRaWAN has been defined to support operation in several world regions. In this PhD thesis, we consider the LoRaWAN physical layer characteristics specified for the European Union (EU), including the use of the 868 MHz band, where three default radio channels are defined: 868.10 MHz, 868.30 MHz, and 868.50 MHz. These channels are characterized by a bandwidth of 125 kHz, use the LoRa modulation, and offer several DRs, from DR0 to DR5, which correspond to 0.3 kbps to 5 kbps, respectively (see Table 1). DR6 and DR7 are optional (the latter is the only one based on the GFSK modulation). In order to provide robustness to the communication, frequency channel hopping is used over the set of radio frequency (RF) channels configured in EDs. Table 1. DRs, SFs, and physical layer bit rates for the EU 868 MHz band channels. DR Modulation SF Bandwidth (kHz) Physical Bit Rate (bit/s) 0 LoRa SF12 125 250 1 LoRa SF11 125 440 2 LoRa SF10 125 980 3 LoRa SF9 125 1760 4 LoRa SF8 125 3125 5 LoRa SF7 125 5470 6 LoRa SF7 250 11,000 7 GFSK 50,000 ETSI regulations establish spectrum access restrictions on the duty cycle, including a duty cycle limitation to less than 1% for the band between 868.0 MHz and 868.6 MHz. LoRaWAN complies with the mentioned duty cycle restriction by introducing an idle interval of suitable duration after the transmission of a message. In class A, message exchanges between an ED and the NS are initiated by the former. As we mentioned earlier, the NS is allowed to transmit only in one of two available receive time windows, called RX1 and RX2 (see Figure 2), which are available after a message transmission by the ED. Therefore, if the NS has to transmit a new downlink message, such a transmission will wait until the next receive window is open. Contributions to the Evaluation and Improvement of LoRaWAN 14 Table 4. Main current consumption details on LoRa/LoRaWAN transceivers used by devices in Table 3. 3.2. Modeling LoRaWAN ED current consumption In this section, we present models of crucial LoRaWAN energy performance parameters such as ED current consumption, ED lifetime, and energy efficiency of data delivery. We assume a class A ED that periodically transmits an uplink data message (e.g., a notification that carries a sensor reading). In the models, we consider the impact of bit errors. For the sake of tractability and clarity, we assume a uniform BER that refers to the residual BER after application of physical layer error correcting techniques, equivalent to the residual BER that corresponds to message loss rate due to non-ideal link quality. The section is divided in two subsections, which offer the aforementioned models for unacknowledged and acknowledged transmission, i.e., the transmission of unconfirmed and confirmed data messages, respectively. We develop the models for all DRs that are mandatory (i.e., from DR0 to DR5), as well as for DR6. 3.2.1. Unacknowledged transmission Our first goal is modeling the average current consumption of an ED in the unacknowledged approach, denoted Iavg_unACK. In order to determine this parameter, we first derive a profile of the different states traversed by the ED, as well as the duration and the current consumed in each state. In order to realistically model the ED behavior, and without loss of generality, we develop the model based on measurements from a real LoRaWAN testbed. We use the MultiConnect mDot platform from Multitech [42] as our reference ED platform for the model, since it is a popular platform, and it is based on the also widely used SX1272 transceiver [46] (see Table 3). While other LoRaWAN ED platforms might exhibit differences with the mDot platform (e.g., due to their internal architecture), we understand that our model captures the main states of a LoRaWAN ED. On the other hand, it must be noted that the mDot platform offers low current consumption decrease (of ~3%, and only in the transmit state) when the voltage applied is reduced from 5 V to 3.3 V, the latter being the lowest voltage that allows the device to operate [42]. However, other platforms may not offer the same current consumption stability as battery voltage decreases over time. Transceiver Current Consumption Sleep Transmit Receive Semtech SX1272 [46] 0.1 µA (max. 1 µA) Min.: 18 mA (7 dBm) Max.: 125 mA (20 dBm) 10.5 or 11.2 mA Semtech SX1276 [47] 0.2 µA (max. 1 µA) Min.: 20 mA (7 dBm) Max.: 120 mA (20 dBm) 10.8, 11.5 or 12.0 mA HopeRF HM-TRLRLF/HFS [48] 2 µA (min. 1.2 µA, max. 3 µA) Min.: 35 mA (13 dBm) Max.: 120 mA (20 dBm) 16 mA (min. 15 mA, max. 18 mA) Microchip RN2483 [35,36] Up to 100-150 µA Min.: 17.3 mA (−4.0 dBm) Max.: 38.9 mA (14.1 dBm) 14.2 mA 3. Modeling the energy performance of LoRaWAN 15 In the measurements, the transmit power of the ED is set to 11 dBm, which is the default value for this parameter. The gateway is a Kerlink LoRa IoT Station platform [49]. Both the ED and the gateway are located in an indoor scenario, where the distance between the ED and the gateway is 2 m. In the measurements, data messages carry a frame payload of the maximum size allowed for each DR in the EU band. Figure 5. Experimental setup for current measurements of the MultiConnect mDot LoRaWAN ED module (on the left) using an Agilent N6705A power analyzer. We assume a periodic behavior for the ED, therefore we model its current consumption during one period. Each period comprises a data message transmission by the ED (including the related procedures required to enable such transmission), otherwise the device is in sleep mode. Note that, for an ED in unacknowledged transmission, current consumption is independent of the BER; that is, regardless of whether channel errors take place in the communication, the ED will consume the same amount of energy for any transmission, since there will not be retransmissions in unacknowledged transmission. Time and current consumption measurement results provided in this section are obtained from several measurements for each tested configuration within a notification period. We found negligible differences within each set of measurements for each configuration. Figure 6 illustrates the current consumption profile of an unacknowledged transmission performed by the MultiConnect mDot ED configured to use DR0 (note that the states traversed and behavior observed are the same for all DRs, except for the duration of some intervals). Table 5 defines and describes the different states involved in an unacknowledged transmission, along with the variables that represent the duration and current consumption of each state. Initially, the ED is in sleep mode, which is characterized by a current consumption three orders of magnitude below that of the rest of states. When the ED starts the procedure to perform the transmission, it first wakes up (state 1), next the radio interface is prepared for activity (state 2), and then the ED transmits the data unit via the radio interface (state 3). After the transmission, the ED disables radio activity and waits (state 4) until it sets the radio into receive mode and remains in the same state for the duration of the first Contributions to the Evaluation and Improvement of LoRaWAN 16 receive window (state 5). Since no incoming preamble is detected, the first receive window is closed, and the ED waits (state 6) until the start of the second receive window. During the latter, the ED radio is turned on for possible incoming data units, until the second receive window is closed due to absence of incoming data (state 7). Note that the shorter duration of the second receive window is due to use of the CAD mechanism (see Section 2.2.1), whereby the ED stops preamble detection much earlier than in the first receive window if no incoming signal is detected. After that, the radio interface is turned off (state 8), a postprocessing interval follows (state 9), and the ED executes a turn off sequence (state 10), prior to returning to the sleep state (state 11). One additional consideration is that we have not identified any specific state due to the internal communication, which takes place via Serial Peripheral Interface (SPI), between the main microcontroller and the radio interface of the mDot platform. This is consistent with the submillisecond latency of payload transmission from the microcontroller to the radio interface, for the range of payload sizes in LoRaWAN, that is typical of SPI. Figure 6. Current consumption profile of a MultiConnect mDot LoRaWAN ED performing an unacknowledged transmission with DR0. The data message transmitted has a FRM Payload size of 51 bytes (i.e., maximum possible size for DR0). Table 5. States, variables and their values for LoRaWAN unacknowledged transmission. State Number Description Duration Current Consumption Variable Value (ms) Variable Value (mA) 1 wake up Twu 168.2 Iwu 22.1 2 radio preparation Tpre 83.8 Ipre 13.3 3 Transmission Ttx (see Table 6) Itx 83.0 4 wait 1st window Tw1w 983.3 Iw1w 27.0 5 1st receive window Trx1w (see Table 6) I1w 38.1 6 wait 2nd window Tw2w Equation (4) Iw2w 27.1 7 2nd receive window Trx2w 33.0 I2w 35.0 8 radio off Toff 147.4 Ioff 13.2 9 Postprocessing Tpost 268.0 Ipost 21.0 10 turn off sequence Tseq 38.6 Iseq 13.3 11 Sleep Tsleep Equation (2) Isleep 45 ×!10−3 3. Modeling the energy performance of LoRaWAN 17 Let TNotif be the time between two consecutive periodic message transmissions performed by the ED, i.e., the notification period. Let Ti and Ii denote the duration and current consumption of state i in Table 5. Iavg_unACK can thus be calculated as shown in Equation (1): 𝐼!"#_%&'() =# 1 𝑇*+,-. & 𝑇-· *!"#"$! -/0 𝐼- # (1) where Nstates is 11 in unacknowledged transmission. Note that Tsleep can be obtained as: 𝑇12334 =𝑇*+,-. −𝑇!5, # (2) where Tact denotes the sum of the durations of all states related with transmission activities, i.e., all states except the sleep interval: 𝑇!5, =𝑇6% +𝑇473 +𝑇,8 +𝑇606 +𝑇7806 +𝑇696 +𝑇7896 +𝑇+.. +𝑇4+1, +𝑇13: # (3) The durations Ttx, Trx1w, and Tw2w are variable and depend on the DR in use. Tw2w actually depends on Trx1w, and can be obtained as: 𝑇696 =𝑅𝐸𝐶𝐸𝐼𝑉𝐸_𝐷𝐸𝐿𝐴𝑌_2−𝑅𝐸𝐶𝐸𝐼𝑉𝐸_𝐷𝐸𝐿𝐴𝑌_1−𝑇7806 # (4) Although Trx2w also depends on the DR, the ED platform in our experiments uses a fixed setting for the second receive window (which corresponds to DR0), and thus the measured value for Trx2w is also constant. We next provide the models to derive Ttx, Trx1w and Trx2w. In order to determine the time needed to transmit a data message via the radio interface, Ttx, we take into acount LoRaWAN procedures, LoRa modulation details and the corresponding regional parameters. Ttx can be expressed in terms of the time required to transmit both the preamble and the physical message, denoted Tpreamble and TPHYMessage, respectively, as follows [31]: 𝑇,8;=#𝑇473!<=23;+𝑇>?@A311!#3 # (5) Tpreamble can be obtained as shown next [46]: 𝑇473!<=23 =𝑇1B< ·(𝑁473 +4.25) # (6) where Npre is the programmed number of symbols to be used by the radio transceiver, the actual physical length of the preamble is (Npre + 4.25) [5], and Tsym is the time of a symbol (in seconds), which depends on the SF and the channel bandwidth (BW, in Hz), as follows [31]: 𝑇𝑠𝑦𝑚 =2𝑆𝐹 𝐵𝑊 ## (7) On the other hand, 𝑇!"#$%&&'(% (in seconds) can be evaluated similarly: 𝑇>?@A311!#3 =𝑇1B< ∗𝑁>?@ # (8) where 𝑁!"# indicates the number of symbols transmitted as the physical message (excluding the preamble), and it can be determined as follows [31]: 𝑁>?@ =8+𝑚𝑎𝑥?𝑐𝑒𝑖𝑙?9IJI·>LJ0M·(N(OP·QR P·(QRO9·TU)D·(𝐶𝑅+4),0D ## (9) In (9), SF corresponds to the spreading factor and can take values from 7 to 12 (which correspond to data rates from DR5 to DR0, respectively); CR denotes the coding rate and can take values from 1 to 4, Contributions to the Evaluation and Improvement of LoRaWAN 18 for 4/5 to 4/8 coding rate, respectively; PL indicates the physical payload length, in bytes. CRC indicates the presence or not of the CRC field in the physical message (CRC is set to 0 if the CRC field is not present; otherwise, CRC is equal to 1); finally, DE, which indicates whether the mechanism to avoid issues regarding drift of the crystal reference oscillator is used or not, takes value 1 for SF12 and SF11 (i.e., it is used for the lowest data rates), and value 0 for the rest of SFs. Equations (5)–(9) can be used to model the duration of both uplink and downlink transmissions (e.g., the latter may correspond to ACKs sent in response to uplink data messages, see Table 6). Table 6. Summary of values for Ttx, Trx1w and Trx2w, along with relevant parameter settings. We have assumed an 8-symbol preamble length, a CR of 4/5 (except for the 20-bit physical header, for which a CR of 4/8 is used), and a bandwidth of 125 kHz (except for DR6, with a bandwidth of 250 kHz, and for DR7, which is based on FSK). Ttx max is obtained by considering the maximum frame payload (FRM Payload) size for each DR, while Ttx min corresponds to the time to transmit a data message that carries no data (e.g., an ACK). In the latter case, the physical frame length is the contribution of the physical header (PHDR) and PHDR_CRC fields, the MAC Header (MHDR), the FHDR and the MIC fields (see Section 2.2 and Section 2.3), leading to a total length of 14.5 bytes plus 8 preamble symbols. Note that the CRC is only present in uplink transmissions (see Section 2.2). For DR7, an additional margin should be added to the Trx1w and Trx2w values to account for possible drifts of the oscillator used for the timer that controls a receive window start. DR SF Tsym Tpreamble Trx1w Trx2w DE FRM Payload Ttx Max Ttx Min Max Min Uplink Downlink (ms) (ms) (ms) (ms) (bytes) (bytes) (ms) (ms) 0 12 32.77 401.41 262.14 33.02 1 51 0 2793.5 991.8 1 11 16.38 200.70 131.07 16.64 1 51 0 1560.6 577.5 2 10 8.19 100.35 98.30 8.45 0 51 0 698.4 288.7 3 9 4.10 50.18 49.15 4.35 0 115 0 676.9 144.4 4 8 2.05 25.09 24.58 2.30 0 242 0 707.1 72.2 5 7 1.02 12.54 12.29 1.28 0 242 0 399.6 41.2 6 7 0.51 6.27 6.14 0.64 0 242 0 199.8 20.6 7 - 0.02 0.48 1.28 1.28 - 242 0 42.4 3.2 We next model the behavior of the hardware module used in our experiments in each receive window when no preamble is detected. For the first receive window, Trx1w can be determined as follows: 𝑇7806 =𝑁W1B< ·𝑇1B< # (10) The ED stays in receive mode for the duration of 𝑁)&*+ symbols. 𝑁)&*+ is 8 symbols for SF = 12 and SF = 11, and 12 symbols for the rest of SFs. For the second receive window, the receiver is active during a fraction of a CAD state (see Section 2.2.1). This fraction has a duration, denoted Trx2w, that can be calculated as shown next: 𝑇7896 =9'(JX9 YZ # # (11) 3. Modeling the energy performance of LoRaWAN 19 After providing the models for determining Ttx, Trx1w and Trx2w, Table 6 summarizes their main values, along with relevant parameter settings used in our ED platform. For Trx2w, we include the corresponding values for the different DR settings possible. However, in our experiments, only the Trx2w value corresponding to DR0 was used. Once all variables required to compute Iavg_unACK are determined, we can calculate the theoretical lifetime of a battery-operated ED that performs unacknowledged transmissions, denoted Tlifetime_unACK, on the basis of the battery capacity, Cbattery (expressed in mA·h), as shown next: 𝑇2-.3,-<3_%&'() =# 𝐶=!,,37B 𝐼!"#_%&'() # (12) Note that the above theoretical ED lifetime calculation assumes an ideal battery with a linear behaviour, whereas the characteristics of a real battery degrade over time. Therefore, the calculated ED lifetime results provided in this document provide an upper bound on the actual ED lifetime that can be expected. Finally, another important performance parameter is the energy cost of data delivery, ECdelivery_unACK, which provides the energy consumed by the ED per each delivered bit of data payload in unacknowledged mode, as shown below: 𝐸𝐶)%,-.%/*_12345 =&𝐼'.(_12345 · 𝑉 · 𝑇267-8 𝐸*𝑙)%,-.%/*_12345, (13) where V denotes the voltage and E[ldelivery_unACK] indicates the expected amount of data successfully delivered by the ED per data frame transmitted. Note that in the previous equation, the numerator computes the energy consumed by the device during Tnotif. Let lpay be the FRM Payload field size (i.e., the amount of data carried in the payload of the data message sent by the ED), and let lData be the total size of the data message, including all headers. Let b denote the BER as introduced in the first paragraph of Section 3.1. Since in unacknowledged transmission there is a single transmission attempt, which may suffer bit errors, E[ldelivery_unACK] is determined as: 𝐸*𝑙)%,-.%/*_12345, =&𝑙!'* ·(1 − 𝑏),!"#" (14) Finally, let us assume that collisions may occur, i.e., an ED message transmission may overlap with messages transmitted by other EDs connected to the same gateway. Let pcoll be the probability that a data message transmitted by an ED collides with at least another message transmission. In order to capture impact of collisions on E[ldelivery_unACK], Equation (14) can be extended as follows: 𝐸G𝑙W32-"37B_%&'()H=#𝑙>!B ·(1−𝑏)2)#"# ·(1−𝑝5+22) # (15) Note that, as already introduced, b corresponds to the residual BER after application of physical layer error correcting techniques, equivalent to the residual BER that corresponds to message loss rate due to non-ideal link quality. In this document, we assume that CR = 4/5, since it is the default CR in LoRaWAN, except for the 20-bit physical header, where CR = 4/8 is used. For CR = 4/5, a parity bit is added to each group of 4 bits from the physical layer message to be transmitted. Assuming that an errorcorrecting code is used for CR = 4/8 (e.g., a Hamming code [50]) and the range of BER values for Contributions to the Evaluation and Improvement of LoRaWAN 20 reasonably useful links, and given the short size of the physical header, we approximate the relationship between b and the physical layer BER, denoted bphy, as shown in the next two equations. Let ploss be the probability that a transmitted message is affected by at least one bit error, and therefore the message is lost, and let lphy_header be the 20-bit header size. Therefore: 𝑝2+11 =#1−(1−𝑏)2)#"# =1−K1−𝑏4[BL\ P·]2)#"#O2*+,_+$#.$/^ # (16) (1−𝑏)2)#"# =K1−𝑏4[BL\ P·]2)#"#O2*+,_+$#.$/^ # (17) 3.2.2. Acknowledged transmission We next model ED average current consumption in the acknowledged transmission approach, based on the corresponding current consumption profile of the same hardware platform as in the previous subsection. In this approach, the ED may behave in two different ways, since the ACK may be transmitted in the first receive window or in the second one. In our model, we consider both options. We initially assume BER = 0, and we subsequently extend the model in order to consider a non-zero BER. Therefore, the average current consumption of an ED in acknowledged mode, denoted Iavg_ACK, can be obtained as shown in the next equation: 𝐼'.(_345 =&𝑝9:-2 · 𝐼'.(_345_9 + 𝑝;:-2 · 𝐼'.(_345_; (18) where Iavg_ACK_1 and Iavg_ACK_2 denote the average current consumption of the ED when the ACK is received in the first and in the second receive window, respectively, and p1win and p2win represent their corresponding probabilities. The LoRaWAN specification offers freedom for network managers and implementers to apply the policy that best suits the requirements of a specific deployment. Therefore, since there is no specific priority by default for the two receive windows, we assume that an ACK may be received by an ED in the first or in the second receive window with the same probability (i.e., p1win = 0.5 and p2win = 0.5). We next derive the models for obtaining Iavg_ACK_1 and Iavg_ACK_2. Figure 7 depicts the current consumption profile of an ED that performs an acknowledged transmission in two different situations: in Figure 7a), the ACK is received in the second window, while in Figure 7b), the ACK is received in the first window. (Note: in our specific scenario, we observed that for DR0-DR3, all ACKs were received in the second window, while for DR4-DR5 all ACKs were received in the first window.) When the ACK is received by the ED in the first window, the number of states involved in acknowledged transmission decreases in comparison with unacknowledged transmission, since the ED does not need to wait for a second receive window (Table 7). On the other hand, duration of the first receive window (Trx1w) and radio off interval (Toff) increase since the ACK needs to be received and subsequently processed. Therefore, Iavg_ACK_1 can be derived by using the same equations used to compute Iavg_unACK (i.e., Equations (1)–(11)), but considering only the states that exist when the ACK is sent in the first receive window (which is equivalent to setting both Tw2w and T2w to 0 in the equations), and the values in Table 7. 3. Modeling the energy performance of LoRaWAN 21 (a) (b) Figure 7. Current consumption profile of a MultiConnect mDot LoRaWAN ED performing an acknowledged transmission: (a) with DR0 (left), (b) with DR5 (right). In the former, the ACK is received by the ED in the second window, whereas in the latter the ACK is received in the first window. The data message transmitted by the ED has a FRM Payload size of 51 bytes (left) and 242 bytes (right), respectively. Table 7. States, variables and their values for LoRaWAN acknowledged transmission when the ACK is sent in the first receive window. State Number Description Duration Current Consumption Variable Value (ms) Variable Value (mA) 1 wake up Twu 169.2 Iwu 22.1 2 radio preparation Tpre 80.4 Ipre 13.7 3 transmission Ttx (see Table 6) Itx 82.8 4 wait 1st window Tw1w 988.4 Iw1w 27.1 5 1st receive window Trx1w (Ttx min in Table 6) I1w 31.8 8 radio off Toff 337.8 Ioff 13.4 9 postprocessing Tpost 272.5 Ipost 20.9 10 turn off sequence Tseq 37.5 Iseq 13.4 11 sleep Tsleep Equation (2) Isleep 45 × 10−3 In order to compute Iavg_ACK_2, we consider that behavior of the ED is similar to that in unacknowledged transmission, since states involved in the corresponding transmission operations are the same, and only two differences can be observed: duration of the second window (T2w) and of the subsequent radio off interval (Toff) are both larger than in the unacknowledged transmission. In fact, the ED needs to stay in the second receive window for the time needed to receive the ACK, and subsequent operations involve processing of the ACK. Therefore, Iavg_ACK_2 can be obtained by means of the same equations used to compute Iavg_unACK (i.e., Equations (1)–(11)), but using the T2w and Toff values that correspond to acknowledged transmission when an ACK is sent in the second receive window (Table 8). Contributions to the Evaluation and Improvement of LoRaWAN 22 Table 8. States, variables and their values for LoRaWAN acknowledged transmission when the ACK is sent in the second receive window. State Number Description Duration Current Consumption Variable Value (ms) Variable Value (mA) 7 2nd receive window Trx2w (Ttx min in Table 6) I2w 38.0 8 radio off Toff 337.8 Ioff 13.4 We next extend the model to compute Iavg_ACK for non-zero BER. We assume that bit errors are uncorrelated. Let Ik denote the average current consumed by the ED when it performs k retransmissions (the last one being successfully acknowledged), since the start of the procedures for the first transmission attempt, until the end of the procedures for the k-th retransmission. Let Iact be the average current consumption due to activities related with transmitting a data message (including retransmissions), i.e., all states except the sleep interval. Iact can be computed as: 𝐼!5, =∑#𝐸[𝐼_]·𝑝_ A'`_NUaN _/b ## (19) where E[Ik] denotes the expected current consumption of an ED when it has performed k data message retransmissions, pk indicates the probability that the ED performs k retransmissions of a message, and MAX_RETR denotes the maximum number of message retransmissions by an ED. The latter is recommended as per the LoRaWAN specification to be set to 7. E[Ik] can be obtained by using the next equation: 𝐸[𝐼_]=𝐼c) _·𝑇c) _+∑P𝐼'()_ac ·K𝑇'()_ac −𝑇7896 -L+#𝐼' -·𝑇' -·𝑝'+𝐼Y -·𝑇Y -·𝑝Y+𝐼( -·𝑇( -·𝑝( 𝑝'+𝑝Y+𝑝(Q _ -/b 𝑇c) _+∑RK𝑇'()_ac −𝑇7896 -L+𝑇' -·𝑝'+𝑇Y -·𝑝Y+𝑇( -·𝑝( 𝑝'+𝑝Y+𝑝(S _ -/0 # (20) In the previous equation, 𝐼<5 - and 𝑇<5 - denote the average current consumption and average duration of the activities related with an i-th acknowledged data transmission attempt which is error-free, and can be computed by using (15) and Tables 6–8. On the other hand, IACK_TO and TACK_TO correspond to the current consumption and the average duration of the ACK_TIMEOUT interval, respectively. As per our measurements, IACK_TO has the same value as Iw1w, whereas ACK_TIMEOUT is a random variable uniformly distributed between 1 and 3 s. Variables 𝐼= - , 𝑇= - and px, where x can be equal to A, B or C, correspond respectively to the average current consumption, duration, and probability of unsuccessful acknowledged data message transmission events defined as follows: A is the event whereby the data message suffers a collision, or it does not suffer a collision but it suffers at least one bit error, and it is equivalent in terms of current consumption and duration to the active part (i.e., all states minus sleep) in unacknowledged transmission; B is the event whereby the data message is successfully received, but the ACK, sent in the first receive window, suffers at least one bit error; and C is the event whereby the data message is successfully received, but the ACK sent in the second receive window, suffers at least one bit error (note that errors in downlink messages can be detected by means of the MIC field). Events B and C are equivalent 3. Modeling the energy performance of LoRaWAN 23 in terms of current consumption and duration to the active part of successful acknowledged transmission with the ACK in the first and in the second window, respectively. Probabilities pA, pB and pC are determined in Equations (21)–(23). Let lData and lAck denote the total size of the data and ACK messages, respectively. Probabilities pA, pB and pC can then be obtained as follows: 𝑝'=𝑝5+22 +(1−𝑝5+22)·(1−(1−𝑏)2)#"#) # (21) 𝑝Y=𝑝06-& · ( 1−𝑝' ) · ( 1− ( 1−𝑏 ) 2012 ) =0.5· ( 1−𝑝' ) · ( 1− ( 1−𝑏 ) 2012 ) # (22) 𝑝(=𝑝96-& · ( 1−𝑝' ) · ( 1− ( 1−𝑏 ) 2012 ) =0.5· ( 1−𝑝' ) · ( 1− ( 1−𝑏 ) 2012 ) # (23) We next determine pk. To this end, we first derive the probability that an ED will send an acknowledged message without performing any retransmissions, p0. Then, p0 can be computed as the probability that the data message will not suffer collisions, and neither the data message nor the ACK will suffer errors: 𝑝b=(1−𝑏)2)#"# ·(1−𝑏)2012 ·(1−𝑝5+22) # (24) Based on p0, pk can be found as the probability that only both message and ACK transmissions that correspond to the k-th message retransmission are successful, as follows: 𝑝_=(1−𝑝b)_·𝑝b # (25) Note that as per Equation (24), when the ED reaches the maximum number of retransmissions, if the last data message retransmission is not successful, the corresponding current consumption is not added to Iact computation in Equation (19). Nevertheless, impact of this inaccuracy is negligible for MAX_RETR = 7 and for practical BER values (e.g., up to 10−3). Therefore, we opt to favor simplicity in our model. On the other hand, for non-zero BER, Tact can be calculated as shown next: 𝑇!5, = & #𝐸 G 𝑇!5,__ H ·𝑝_ A'`_NUaN _/b # (26) where E[Tact_k] can be determined by using the next equation: 𝐸 G 𝑇!5,__ H =𝑇c) _+ &RK 𝑇'()_ac −𝑇7896 - L +𝑇' -·𝑝'+𝑇Y -·𝑝Y+𝑇( -·𝑝( 𝑝'+𝑝Y+𝑝( S _ -/0 # (27) Based on Equations (18)–(23), Iavg_ACK can be obtained by considering the active interval and the sleep interval over the notification period, TNotif, as: 𝐼!"#_'() =𝐼!5, ·𝑇!5, +𝐼12334 ·(𝑇&+,-. −𝑇!5,) 𝑇&+,-. # (28) The previous equation can be used to calculate the theoretical lifetime (i.e., an upper bound on the actual lifetime) of a battery-operated ED that performs acknowledged transmissions, denoted Tlifetime_ACK, on the basis of the battery capacity, Cbattery (expressed in mA·h), and Iavg_ACK, as shown next: 𝑇2-.3,-<3_'() =#𝐶=!,,37B 𝐼!"#_'() # (29) Contributions to the Evaluation and Improvement of LoRaWAN 30 in the figure for the sake of illustration clarity. Impact of the payload size is noticeable only when TNotif is low. For DR0 and TNotif = 5 min, ED lifetime ranges from 0.27 up to 0.41 years (i.e., a ~52% relative difference), for the payload sizes considered. For DR5 and TNotif = 1 min, ED lifetime ranges from 0.18 to 0.26 years (i.e., a relative difference of ~49%). However, impact of frame payload size decreases when TNotif increases because sleep state, and thus its current consumption, becomes dominant. For example, for a TNotif = 60 min and DR0, ED lifetime values range from 2.17 to 2.90 years (i.e., 33.4% difference), and from 3.90 to 4.44 years (i.e., a 13.78% difference) for DR5. Figure 13. ED lifetime in unacknowledged transmission, as a function of TNotif, and for different DR settings. Next, we evaluate impact of BER on ED lifetime (Figure 16). Similarly to the observations made in Section 3.3.2, ED lifetime may decrease by up to one order of magnitude for DR5 and for the range of BER values considered. Impact of BER on ED lifetime is lower for DR0, and it decreases with TNotif, since ED lifetime tends asymptotically to the lifetime of an always-sleeping ED as TNotif increases. Finally, we also study the impact of pcoll on ED lifetime (Figure 17). Collisions may significantly reduce ED lifetime (e.g., by one third for TNotif = 30 min and pcoll = 0.3), and have a greater impact on ED lifetime for DR5, as expected from the average current consumption analysis in Section 3.3.2. 3.3.4. Energy cost of data delivery In this subsection, we evaluate the last considered performance parameter, i.e., the energy cost of data delivery for both unacknowledged and acknowledged transmission. We next apply Equations (13) and (14) and (30) and (31) to determine ECdelivery_unACK and ECdelivery_ACK, respectively. We assume a battery voltage of 3.6 V. Figure 18 provides the energy cost of data delivery for the unacknowledged approach, as a function of TNotif and the DR used, for BER = 0. 0 1 2 3 4 5 6 0,1 110 100 1000 Lifetime (years) Notification period (min) DR0, No ACK DR1, No ACK DR2, No ACK DR3, No ACK DR4, No ACK DR5, No ACK DR6, No ACK 3. Modeling the energy performance of LoRaWAN 31 For a given DR, and for BER = 0, the relative difference in energy cost of data delivery between acknowledged and unacknowledged transmission is the same as the difference in terms of current consumption analyzed in Section 3.3.2. However, such difference is only significant for low notification periods, and therefore it is not graphically visible in Figure 16, therefore the figure serves for both unacknowledged and acknowledged approaches. Figure 14. ED lifetime in acknowledged and unacknowledged transmission, as a function of TNotif, and for different DR settings, BER = 0 and pcoll = 0. DR2, DR3 and DR5 are not shown in the figure for the sake of illustration clarity. Figure 15. Comparison of the ED lifetime with 1-byte and maximum-sized payload data message in acknowledged transmission, as a function of TNotif, for BER = 0, pcoll = 0, and for different DR settings. Only DR0 and DR5 are shown in the figure for the sake of illustration clarity. 0 1 2 3 4 5 6 0,1 110 100 1000 Lifetime (years) Notification period (min) DR6, No ACK DR6, ACK DR4, No ACK DR4, ACK DR1, No ACK DR1, ACK DR0, No ACK 0 1 2 3 4 5 6 0,01 0,1 110 100 1000 Lifetime (years) Notification period (min) DR0 ACK, 1-byte payload DR5 ACK, 1-byte payload DR0 ACK, max payload DR5 ACK, max payload Contributions to the Evaluation and Improvement of LoRaWAN 32 Figure 16. Impact of BER on ED lifetime in acknowledged transmission, as a function of TNotif, for different DR settings, and for pcoll = 0. Figure 17. Impact of pcoll on ED lifetime in acknowledged transmission, as a function of TNotif, for different DR settings, and for BER = 0. As shown in Figure 18, for BER = 0, the energy cost of data delivery follows a linear trend as a function of TNotif. As it has been previously shown in Figures 8 and 9, average current consumption, and therefore energy consumption becomes asymptotically constant as a function of TNotif. Therefore, as TNotif increases, the energy consumed by the ED increases linearly with time, while the number of delivered bits remains constant. Because we are considering the maximum frame payload size allowed by each DR, note that the slope of the curves for DR0-DR2 is the same, since these three DRs allow the same maximum frame payload size (i.e., 51 bytes), the slope for DR3 is lower, since the maximum frame payload size is greater (i.e., 115 bytes), and finally the slope for DR4, DR5 and DR6 is the lowest, since for these DRs the maximum frame payload size is the largest supported by LoRaWAN (242 bytes). DR0 exhibits slightly greater energy cost of data delivery than DR1, due to the lower bit rate of DR0, which leads to a greater data message and ACK transmit time, as well as receive window duration. The same reasoning applies to 0 1 2 3 4 5 6 110 100 1000 Lifetime (years) Notification period (min) DR0, BER=10⁻³ DR0, BER=10⁻⁴ DR0, BER=10⁻⁵ DR5, BER=10⁻³ DR5, BER=10⁻⁴ DR5, BER=10⁻⁵ 0 1 2 3 4 5 6 110 100 1000 Lifetime (years) Notification period (min) DR0, Pcoll=0.3 DR0, Pcoll=0.2 DR0, Pcoll=0.1 DR0, Pcoll=0 DR5, Pcoll=0.3 DR5, Pcoll=0.2 DR5, Pcoll=0.1 DR5, Pcoll=0 3. Modeling the energy performance of LoRaWAN 33 the comparison of the energy cost of data delivery of DR1 and DR2. The energy cost of data delivery for DR4 is ~19% and ~40% greater than the one for DR5 and for DR6, respectively, for the lowest notification period, and decreases down to ~1% for both DR5 and DR6 for the highest notification period considered (note that these differences are not visible in Figure 18). We next evaluate the upper bound on the energy cost of data delivery by considering a frame payload size of 1 byte. Results are shown in Figure 19, along with the ones obtained for a maximum-sized frame payload. As it can be seen, for a given notification period and DR, the energy cost per delivered payload bit for a 1-byte frame payload is roughly two orders of magnitude greater than the one obtained for a maximum-sized frame payload. We then study the impact of non-zero BER on the energy cost of data delivery, assuming a maximum-sized payload for DR0, and the same payload size for DR5, and considering also unacknowledged and acknowledged data message transmission (see Figure 20). Acknowledged transmission leads to a greater energy cost, by at least one order of magnitude, due to the additional energy consumption of retransmissions, in comparison with the unacknowledged approach. On the other hand, BER has a greater impact on DR5 than on DR0, since the former transmits (and may retransmit) data messages at a higher bit rate, leading to lower energy consumption, which makes retransmissions, and their related overhead, energy-expensive. We next focus on how collisions influence the energy cost of data delivery (see Figure 21). As expected, collisions increase the energy required per delivered payload bit. In acknowledged transmission, this effect is emphasized, since retransmissions are needed. For example, for DR5, the energy cost of data delivery in acknowledged transmission for Tnotif = 1 min increases by 60% when pcoll increases from 0.1 to 0.3, while in unacknowledged mode, the energy cost per delivered bit increases by 12.8%. Figure 18. Energy cost of data delivery as a function of TNotif, for different DR settings, and for BER = 0 and pcoll = 0. Results for DR4, DR5 and DR6 overlap in the figure. 0 1 2 3 4 5 6 0200 400 600 800 1000 1200 1400 Energy cost (µJ/bit) Notification period (min) DR0 DR1 DR2 DR3 DR4-DR6 Contributions to the Evaluation and Improvement of LoRaWAN 34 Figure 19. Energy cost of data delivery as a function of TNotif, for different DR settings, BER = 0, pcoll = 0, and for 1-byte payload and maximum-sized payload that corresponds to each DR. Figure 20. Impact of BER on the energy cost of data delivery, as a function of TNotif, for both unacknowledged and acknowledged transmission, for pcoll = 0, and for DR0 and DR5. Figure 21. Impact of pcoll on the energy cost of data delivery, as a function of TNotif, for both unacknowledged and acknowledged transmission, for BER = 0, and for DR0 and DR5. Finally, we analyze and discuss the impact of payload (and thus, message) size on the energy cost per delivered bit in a dense LoRaWAN network, where collisions may occur. Assuming that EDs perform 0,01 0,1 1 10 100 1000 0,01 0,1 110 100 1000 10000 Energy cost (µJ/bit) Notification period (min) DR0, 1-byte payload DR1, 1-byte payload DR2, 1-byte payload DR3, 1-byte payload DR5, 1-byte payload DR6, 1-byte payload DR0, max payload DR1, max payload DR2, max payload DR3, max payload DR5, max payload DR6, max payload 0,1 1 10 100 110 100 1000 Energy cost (µJ/bit) Notification period (min) No ACK, DR0, BER=10⁻³ No ACK, DR0, BER=10⁻⁴ No ACK, DR5, BER=10⁻³ No ACK, DR5, BER=10⁻⁴ ACK, DR0, BER=10⁻³ ACK, DR0, BER=10⁻⁴ ACK, DR5, BER=10⁻³ ACK, DR5, BER=10⁻⁴ 0,1 1 10 100 110 100 1000 Energy cost (µJ/bit) Notification period (min) ACK, DR0, Pcoll=0.3 ACK, DR0, Pcoll=0.2 ACK, DR0, Pcoll=0.1 No ACK, DR0, Pcoll=0.3 No ACK, DR0, Pcoll=0.2 No ACK, DR0, Pcoll=0.1 ACK, DR5, Pcoll=0.3 ACK, DR5, Pcoll=0.2 ACK, DR5, Pcoll=0.1 No ACK, DR5, Pcoll=0.3 No ACK, DR5, Pcoll=0.2 No ACK, DR5, Pcoll=0.1 3. Modeling the energy performance of LoRaWAN 35 acknowledged transmissions, network behavior can be modeled by an Aloha access protocol, as an approximation [5]. Under these conditions, the energy cost per delivered bit can be computed as the result of dividing the energy cost of a message transmission (including its retransmissions) by the number of payload bits carried. The energy consumed in a message transmission (including retransmissions) is roughly proportional to the number of transmission attempts per message. In fact, the amount of energy consumed in each retransmission (which includes the energy consumed over ACK_TIMEOUT) is much larger than the energy consumed during the actual transmission state, regardless of the message size, with an accuracy that increases with the DR since transmission time decreases, and with the load offered to the network. On the other hand, in Aloha, the expected number of transmission attempts per message is e2G, where G is the total load offered to the network, and thus the expected energy consumed to deliver a message is a·e2G, where a is the total amount of energy consumed in each transmission attempt. Note that G is proportional to the message size, i.e., lpay + lhead, where lhead denotes the total size of the message headers. Therefore, the energy cost of data delivery under the described conditions can be approximated by (32): 𝐸𝐶)%,-.%/* ≈𝑎 · 𝑒;> 𝑙?'* =𝑎 · 𝑒@·(,$"%C,&'"() 𝑙?'* = &𝛼 · 𝑒@·,$"% 𝑙?'* (32) where α is a constant expressed in Joules, that can be computed by using (33), and k is a constant expressed in bit−1, which depends on the transmission bit rate and on the total message rate offered to the network. 𝛼 = 𝑎 · 𝑒@·,&'"( (33) Note that the model presented cannot accurately capture the DR decrease mechanism for retransmissions (see Section 2.3), when used, since Aloha assumes all packet transmissions have the same duration. Nevertheless, it allows to qualitatively capture behavior of the energy cost of data delivery as a function of packet size. As shown in Figure 22, impact of payload size on the energy cost per delivered bit depends on the value of k. All curves follow a “U” shape with an optimal payload size that minimizes the energy cost per delivered bit. The “U” shape of the curves can be explained, on the one hand, by the fact that very large messages will lead to high energy cost per delivered bit due to a high number of collisions. On the other hand, messages with very small payload will also lead to high energy cost per delivered bit, because while the number of collisions will be relatively low, the energy cost will be relative to a low amount of delivered bits. As k increases, impact of collisions becomes dominant and the payload size that minimizes the energy cost per delivered bit decreases. Note that for sufficiently low, or sufficiently high, k values, the optimal payload size falls out of the region of valid payload sizes. For example, for a high enough k (e.g., k = 0.1), the corresponding energy cost per delivered bit curve in Figure 21 is a function that grows steadily with payload size. Contributions to the Evaluation and Improvement of LoRaWAN 36 Figure 22. Impact of message payload, lpay, on the energy cost of data delivery, assuming a dense LoRaWAN network, for, α = 0.4 J and DR = 5. 3.4. Conclusions In this first study, we have modeled the energy consumption of a class A LoRaWAN ED transmitting data messages periodically, considering impact of unacknowledged and acknowledged transmission, DRs, frame payload size and BER. Performance parameters have been ED average current consumption and lifetime, and energy cost of data delivery. The models have been developed based on measurements performed on prevalent LoRaWAN hardware. For BER = 0, acknowledged transmission reduces LoRaWAN ED average current consumption. This happens because an ACK may be sent in the first receive window, whereas unacknowledged transmission involves two receive windows and a larger average current consumption than a transmission ACK in the first receive window. Note that the quantitative difference between energy consumption in acknowledged and unacknowledged transmission for BER = 0 may be platform-specific. In fact, in contrast with the behavior observed with the platform used in this work, the ED hardware platform might stay in a low consumption mode (e.g., such as the sleep state) during the interval between the first and the second receive windows. On the other hand, for a non-negligible BER, acknowledged transmission leads to greater current consumption than unacknowledged transmission, due to message retries. For a notification period of 5 min, DR5 leads to a current consumption lower than that of DR0 by a maximum factor of 2.8, whereas a 1-byte frame payload size reduces current consumption by up to a maximum factor of 1.59. For the same notification period, using DR6 (which is not mandatory as per the LoRaWAN specification), current consumption decreases by a factor of 3.18, compared to using DR0. Nonzero BER up to 10−3 may increase current consumption by up to one order of magnitude in acknowledged transmission. Current consumption differences due to the settings considered tend to decrease with the notification period, since sleep current consumption then becomes dominant. 0,001 0,01 0,1 1 110 100 Energy cost (J/bit) Payload size (bytes) k=0.1 k=0.05 k=0.01 k=0.003 3. Modeling the energy performance of LoRaWAN 37 An ED running on a 2400 mAh battery and sending one message every 5 min can achieve a 1-year lifetime. As the notification period increases, the theoretical ED lifetime tends asymptotically to roughly 6 years under the conditions considered. However, from our state of the art analysis, that at the time of this study LoRaWAN hardware is not as well optimized as that of other low-power technologies. In the latter, sleep current in the order of (or even below) 1 µA is common, which allows multiyear device lifetimes with a button cell battery, of 1 order of magnitude less capacity than the one considered in this study. In contrast, current batteries used for LoRaWAN hardware have a larger size and weight, which in turn has an impact on the physical dimensions and weight of current LoRaWAN devices and limits applicability of current LoRaWAN devices for domains where such dimensions are relevant, such as wearables. Finally, the energy cost per delivered bit of maximum-sized frame payload transmission is roughly two orders of magnitude lower than the one obtained for the short payload of 1 byte per frame. An ED operating as a sensor will thus benefit significantly from accumulating readings and sending them at the highest notification period possible. For non-zero BER up to 10−3, acknowledged transmission increases the energy cost of data delivery by up to roughly one and two orders of magnitude, for DR0 and DR5, respectively. The influence of the frame payload over the energy cost, for BER = 0, will be studied in depth in Chapter 5, by means of simulations. 4. The SF12 Well in LoRaWAN: problem and ED solutions Since the initial specification of LoRaWAN was published, many researchers have devoted their efforts to investigating the performance of this technology [1,5,51-56]. However, LoRaWAN network performance may become compromised as the number of connected ED increases, even for a relatively low number of such devices. LoRaWAN offers optional link-layer reliability, based on link-layer ACKs and retransmissions (see Section 2.3). When ACKs are not received by a sender after transmitting a frame in reliable mode, the sender typically tends to reduce its physical layer bit rate. However, this reaction increases uplink and, even worse, downlink channel utilization. As a result, network performance may degrade steadily into congestion collapse. In this chapter, we identify and illustrate a problem which we call the Spreading Factor 12 (SF12) Well in a range of scenarios, and evaluate a number of solutions to counter it. We show that it is possible to mitigate it, maintaining good performance as the offered load increases by using different SF management techniques. In order to carry out the study, we developed and used a simulator called Advanced Framework for LoRa (AFLoRa), a LoRaWAN simulation environment that uses the FLoRa simulator [57] as a basis, albeit with significant enhancements and additions. As a side-contribution of this chapter, we offer the simulator publicly [58]. The present chapter is organized as follows. First, in the next section, we review related work. Section 4.2 illustrates the SF12 Well problem in different scenarios in terms of the offered load and the ratio of acknowledged traffic. Section 4.3 proposes and evaluates a number of techniques intended to counter the SF12 Well problem, showing their performances and trade-offs. Finally, Section 4.4 provides the main remarks from this work. 4.1. Related work Previous work has pointed out that confirmed traffic in LoRaWAN may be impaired by duty cycle restrictions that limit downlink capacity, rendering gateways as bottlenecks that cause packet delivery ratio (PDR) decrease [52-54,59-61]. However, the consequent problem of congestion collapse remained unexplored before this writing. Contributions to the Evaluation and Improvement of LoRaWAN 46 (a) (b) Figure 27. Evolution of cumulative packet drops at the EDs due to duty cycle restrictions over time. (a) Lowest load: 30 CEDs, using an EDL of 0.9 packets/h. (b) Highest load: 100 CEDs, using an EDL of 18 packets/h. Finally, Figure 28 presents the instantaneous PDR over time. Note that the instantaneous PDR corresponds to the ratio of packets received by the NS at the application layer divided by the total number of application-layer packets generated by the EDs within a relatively small interval (of 4 h and 0.2 h in Figures 28a and 28b, respectively) that starts at that given time. After the WFT, this performance parameter reflects the steady-state network performance. As aforementioned, in both scenarios, all the CEDs use SF12 after the WFT. However, for low loads, the PDR value after the WFT is near 100%, while in heavy load conditions the PDR is very low, around 10%. In that case, the SF12 Well problem has a dramatic impact on network performance. Figure 29 shows the WFT as a function of the node load. As the offered traffic decreases, the WFT increases quickly. The greatest WFT value is obtained for 30 CEDs and EDL = 0.9 packets/h, with an average WFT equal to 65.3 days. On the other hand, the lowest WFT value is found for 100 CEDs and the EDL = 18 packets/h, with an average WFT equal to 76.7 min (and a maximum value of 93.9 min). Therefore, the WFT depends greatly on the number of CEDs and on the EDL. As the EDL decreases, the mean WFT increases very quickly (note the logarithmic scale for the vertical axis in Figure 29). The WFT differences for the different numbers of CEDs increase as well. 4. SF12 well in LoRaAWAN: problem and ED solutions 47 (a) (b) Figure 28. Instantaneous PDR after WFT. (a) Lowest load: 30 CEDs, using an EDL of 0.9 packets/h. (b) Highest load: 100 CEDs, using an EDL of 18 packets/h. We next study the impact of the EDL and the number of CEDs on the PDR (see Figure 30). It can be observed how such an impact is significant. For low EDL values, as the EDL increases, the PDR decreases quickly, especially for a high number of CEDs. However, the decrease slows down as the EDL increases. This behavior can be understood by looking at the PDR results corresponding to the CED and UED transmissions (see Figure 31 and Figure 32, respectively). While in the former the PDR exhibits a steady degradation with the EDL, the latter shows only a slight PDR decrease with the EDL, even for high EDL values. Figure 29. Mean and standard deviation of WFT as a function of EDL, for several numbers of CEDs. Contributions to the Evaluation and Improvement of LoRaWAN 48 Figure 30. PDR vs EDL, from all EDs. Figure 31 also shows that the EDL has a dramatic effect on the PDR, while the number of CEDs generally has a lower influence on the PDR. However, for medium loads, the number of CEDs becomes more relevant. This result can be attributed to the higher impact of the unconfirmed transmissions at high loads when the number of CEDs is low. On the other hand, the high PDR for unconfirmed transmissions shown in Figure 32 indicates that, in contrast with the CEDs, the UEDs are not affected by network congestion, even under a high UED load. Furthermore, the UEDs’ PDR is almost not influenced by the CED transmissions or by the number of CEDs. This behaviour is due to the constant use of SF7 (i.e., the initial SF value for all the EDs in our considered scenarios) by the UEDs, which yields the lowest possible frame ToA. Figure 31. PDR for CEDs vs EDL. 4. SF12 well in LoRaAWAN: problem and ED solutions 49 Figure 32. PDR for EDs in unconfirmed transmission mode vs EDL. In order to further analyze the reasons for the network PDR behavior, we next focus on the additional performance parameters. Figure 33 illustrates the number of MAC frame collisions over the total number of transmitted frames, along with the number of dropped packets due to duty cycle restrictions over the total number of packets intended to be transmitted by the EDs and by the gateway, respectively, in the considered scenarios. Note that in those scenarios, the downlink traffic forwarded by the gateway comprises only the ACKs sent by the NS. As load increases, the number of collided frames and dropped packets also increases quickly for low loads. However, for a high EDL, this increase slows down, even becoming a decrease, except for scenarios with 100 CEDs (see Figures 33a and 33c). There are two main reasons for this decrease. First, as the CEDs increase their SF, the frame ToA increases, therefore the number of duty cycle losses increases. Consequently, for high EDL the CED collisions do not increase as much as for the low EDL because there is a reduced number of CED transmissions. Second, the UED transmissions contribute a low number of collisions due to their use of SF7, as discussed earlier. On the other hand, the packet drops at the EDs affect the PDR for both types of EDs, while packet drops at the gateway have a negative effect, mainly on the CEDs’ PDR. All these behaviors have a great impact on the PDR (see Figure 30), leading to a PDR decrease with the EDL that slows down as the EDL increases. The observations from Figure 33 are also relevant to understand the possible impact of the number of channels on the SF12 Well problem. For a greater number of channels (e.g., 8 channels), a lower number of data frame collisions and ED packet drops, and thus a PDR increase, is expected. While the number of ED retries might appear to decrease as a result, thus delaying the SF12 Well problem, another consequence of the PDR increase is a greater number of ACKs to be sent by the gateway, increasing packet (i.e., ACK) drops at the gateway. The EDs awaiting the ACKs that will not be received will anyway perform retries, and increase their SF, for the corresponding packets (even if such packets have actually been successfully delivered). Contributions to the Evaluation and Improvement of LoRaWAN 50 (a) (b) (c) Figure 33. (a) Collision ratio at the gateway. (b) Packet drop ratio at EDs due to duty cycle restrictions. (c) Packet drop ratio at the gateway due to duty cycle restrictions. 4.3. ED-based solutions The analysis in the previous section has shown that all the CEDs evolve to use SF12 at a given WFT in the scenarios considered. Thereafter, the EDs use the worst SF configuration in terms of network congestion. In this section, we propose three alternative SF management techniques based on simple changes to how the SF parameter is managed by the ED. Then, we present and discuss an evaluation of these solutions. 4.3.1. Proposed solutions As explained in Section 2.3, LoRaWAN devices typically increase the SF after two consecutive ACKs not received. This behavior will henceforth be referred to as SF Mode 0 (SFM0). As alternatives, we introduce three new SF parameter management techniques, namely SF Mode 1 (SFM1), SF Mode 2 (SFM2), and SF Mode 3 (SFM3). They are defined as follows: 1. SFM1. SF7, which corresponds to the greatest DR value, is always used. No SF change is conducted even if the data or the ACK frames are lost. 2. SFM2. This technique adds an extra step to SFM0. When the SF reaches the SF12 value, if the corresponding ACK is not received after two transmission opportunities, the SF is reset to SF7. The rationale for this approach is that after unsuccessful transmission using SF12 it may be better to switch to SF7 as congestion might be the reason for the frame losses. Hence, SFM2 leads to a cyclic use of all the SF values. 3. SFM3: The basis for this technique is also SFM0. However, with this technique, if an ACK is received, the ED will decrease its SF value. Hence, this option brings the opportunity to increase the DR, with an expectation to reduce the frame ToA and thus reduce network congestion. 4. SF12 well in LoRaAWAN: problem and ED solutions 51 Note that we consider SFM1 as a benchmark for our simulation scenarios as it minimizes the frame ToA. Obviously, it cannot be considered a general solution (i.e., for any type of LoRaWAN network) because its performance will be severely affected for high numbers of CEDs, due to collisions, or if the radio link quality is not good enough. 4.3.2. Evaluation In this subsection, we evaluate the performance of the presented alternative SF management techniques (i.e., SFM1, SFM2, and SFM3) in the same conditions considered for SFM0 in Section 4.2. We focus on the network PDR as the main performance metric (Figure 34). However, in order to better understand the network performance and behavior, we also study the number of collisions (Figure 35), the number of losses due to duty cycle restrictions at both the ED (Figure 36) and the gateway (Figure 37), and the distribution of the SF values used for each mode, and for each considered number of CEDs (Figure 38). Figure 34 illustrates the total PDR for traffic including confirmed and unconfirmed transmission modes, as a function of the EDL. SFM1 (see Figure 34a) yields the highest PDR among the considered SF management modes, even in high load conditions. For SFM1, the PDR is always above 80% in the considered scenarios. SFM1 also offers the best behavior in terms of losses due to duty cycle restrictions (see Figure 36a and Figure 37a). This best performance under congestion is due to the fact that SF7 leads to the minimum ToA, which minimizes channel utilization and the impact of the duty cycle limitation on network performance. This occurs despite the fact that the collision ratio of SFM1 is not the best among the considered SF management techniques, except for a very low packet load (see Figure 35). The observed collision ratio is a consequence of the approach in SFM1 based on always using the same SF value (i.e., SF7), which precludes exploiting the orthogonality that stems from using different SF values. It is interesting to compare the high PDR achieved by using SFM1 (Figure 34a), with the low PDR obtained by the CEDs when using SFM0 (Figure 31) for the same range of EDL values and number of CEDs. The low PDR of SFM0 is mainly due to the use of high SF values (equal to SF12 for all CEDs after the WFT), which lead to high ToA values and produce a high number of collisions at the gateway (Figure 33a) and packet drops at the EDs (Figure 33b). The gateway is also unable to transmit all the ACKs to the corresponding CEDs due to duty cycle restrictions (Figure 33c), leading to unnecessary retries by those CEDs which increase the offered load and contribute to the decrease of the PDR. In contrast, in SFM1 all nodes use SF7, which minimizes the ToA and yields a high PDR. However, we hypothesize that there exists a higher number of CEDs and/or traffic load that will create a low PDR problem equivalent to the one found for SFM0, regardless of the SF values used by the EDs. Contributions to the Evaluation and Improvement of LoRaWAN 52 (a) (b) (c) Figure 34. PDR vs EDL for (a) SFM1, (b) SFM2, and (c) SFM3. In contrast with the SFM1 behavior, SFM3 leads to a quick PDR decrease as the offered load increases, especially for a high number of CEDs (Figure 34c). This is due to an increase in the number of CEDs that use SF12 or an SF value close to this one (see Figure 38i), leading to greater ToA values and contributing to higher collision probability (see Figure 35c), which results in low PDR values. On the other hand, for a low number of CEDs, the PDR obtained with SFM3 tends to be the highest among all of the considered SF management modes. This is, again, related to the distribution of the SF values used by the CEDs: they mainly use SF7 for a low number of CEDs, even for high EDL (see Figure 38i). This SF value distribution leads to a low collision ratio (see Figure 35c) and a very low number of losses due to duty cycle restrictions, at both the EDs (see Figure 36c) and the gateway (see Figure 37c). Finally, SFM3 offers a higher PDR compared with SFM0 by a factor of up to 2.44, within the study conditions. (a) (b) (c) Figure 35. Frame collisions at the gateway vs EDL for (a) SFM1, (b) SFM2, and (c) SFM3. 4. SF12 well in LoRaAWAN: problem and ED solutions 53 (a) (b) (c) Figure 36. Packet losses due to duty cycle restrictions at the EDs vs EDL for (a) SFM1, (b) SFM2, and (c) SFM3. SFM2 leads to intermediate PDR results (see Figure 34b) because it produces a distribution of SF values used by the CEDs that is rather uniform (see Figure 38b, 38e and 38h). Only a significant SF value distribution difference exists for a very low EDL, but even in this case, the probability of using the most likely SF value (i.e., SF7) is below 41% for all numbers of the CEDs considered (see Figure 38b). The more even distribution of SF values achieved by SFM2 (see Figure 38h) exploits the orthogonality of different SF values and leads to less collisions than the rest of the considered SFMs, especially for a high load (see Figure 34). However, the number of losses due to duty cycle restrictions of SFM2 is greater than the one obtained for SFM1 (at both the CEDs and the gateway), and, for high EDL and a high number of CEDs, it is also greater than the SFM3 one, regarding losses at the gateway (see Figure 36b and Figure 37b, respectively). Such behavior of SFM2 is due to its tendency to lead to greater frame ToA, because of its greater probability of using greater SF values than those of SFM1 and SFM3 (regarding the latter, one exception is the high EDL and the high number of CEDs). Nevertheless, SFM2 yields a PDR increase, compared with SFM0 (Figure 30), by a factor of up to 4.7 for the range of scenarios considered. (a) (b) (c) Figure 37. Packet losses due to duty cycle at the gateway vs EDL for (a) SFM1, (b) SFM2, and (c) SFM3. Contributions to the Evaluation and Improvement of LoRaWAN 54 On the other hand, comparing losses due to duty cycle restrictions at the gateway for SFM0, SFM2, and SFM3 schemes (Figures 33c, 37b and 37c), we can appreciate an increase in those losses for the mentioned alternative SFMs. For SFM0, the lower number of packet (ACK) drops at the gateway is due to a lower PDR (which is due to a greater number of packet losses due to collisions, and the duty cycle restrictions at the EDs) and thus a lower number of ACKs to be sent. For the alternative SF management schemes, the PDR is greater than for SFM0, which increases the ACK traffic, and the number of packet drops at the gateway. Therefore, the gateway becomes a limitation for the alternative SF management techniques. From the above analysis, it can be highlighted that any of the considered alternative SF management techniques allows the avoidance of the SF12 Well and improves the network PDR. SFM1 offers the best performance in terms of the PDR and packet losses due to duty cycle restrictions, although, as aforementioned, this SF management technique is only included in the evaluation as a benchmark. Both SFM2 and SFM3 outperform SFM0 but offer different trade-offs. In terms of collisions at the gateway, SFM2 is the best option because it presents a more even SF value distribution. However, for low loads, SFM3 offers a lower number of duty-cycle-induced packet drops, due to a higher fraction of devices using low SF values. In consequence, SFM2 tends to offer a greater PDR than SFM3 for high EDL and a high number of CEDs, whereas SFM3 yields a higher PDR than SFM2 for a low EDL and a low number of CEDs. 4. SF12 well in LoRaAWAN: problem and ED solutions 55 SFM1 SFM2 SFM3 (a) (b) (c) 0.9 packets/h (d) (e) (f) 3.6 packets/h (g) (h) (i) 18 packets/h Figure 38. CED SF values distribution when SF mode is (a, d, and g) SFM1; (b, e, and h) SFM2; and (c, f, and i) SFM3 and for (a, b, and c) EDL equal to 0.9 packets/h, (d, e and f) 3.6 packets/h, and (g, h and i) 18 packets/h. 4.4. Conclusions In this chapter, we identified and characterized by simulation a formerly unexplored LoRaWAN network condition, which we call the SF12 Well. This phenomenon may arise due to the presence of even a relatively low number of CEDs, which will tend to increase their SF and thus the number of collisions and packet drops due to duty cycle constraints. In consequence, the SF12 Well may significantly degrade Contributions to the Evaluation and Improvement of LoRaWAN 62 For SF12 (see Figures 39b, 40b and 41b), the EPB is greater than that obtained for SF7, due to the greater transmission time. The latter also impacts other performance parameters, such as the number of collisions and frame drops, which further increase energy consumption. However, EPB decreases steadily with packet size within the range of valid packet sizes, and thus, it does not show the “U” shape found in some cases for SF7. In order to determine the reasons behind the obtained EPB results as a function of packet size, we next focus on the following additional performance parameters from the same evaluated scenarios: i) collision ratio (Figure 42), ii) the downlink frame drop ratio at the gateway –hereinafter referred to as gateway drop ratio– (Figure 43), iii) the packet drop ratio at EDs –hereinafter referred to as ED drop ratio– (Figure 44), iv) the frame retransmission ratio (Figure 45), and v) PDR (Figures 46 and 47). In each figure, we present results for two TP values: 1000 s, and 4000 s. (a) (b) Figure 41. EPB as a function of packet size, for TP = 4000 s: (a) SF7, (b) SF12. First, we analyze the performance for SF7, and N values of 1000 and 2000 (i.e., the SF and N values for which EPB as a function of packet size shows an “asymmetric U” shape), in terms of collision ratio, ED drop ratio, and gateway drop ratio. For low packet sizes, the collision probability and ED drop ratio are relatively low (see Figure 42a and 8a), while the gateway drop ratio is high (see Figure 43a). The latter is due to the duty cycle restriction (i.e., 1% in the used frequency band in the downlink channel) and the high number of EDs (all of them CEDs) that require ACKs in response. In consequence, EPB is high due to a high number of unnecessary ED retransmissions, and also due to the low packet size itself. When packet size increases, initially, EPB decreases due to the dominant effect of a greater amount of delivered bits over constant energy overheads. However, the collision ratio increases quickly, and the ED drop ratio also increases; in consequence, the gateway drop ratio decreases, since the NS receives a 5. Energy efficiency-optimal packet size in LoRaWAN 63 lower number of frames, and therefore it sends a lower number of ACKs in response to the EDs. For high packet sizes, ED retransmissions (Figure 45a) increase due to the very high collision ratio and ED drop ratio, leading to an EPB increase. However, such increase is mitigated by the greater packet size, leading to a low slope and an "asymmetric U” shape for N = 1000 and N = 2000, and SF7, as shown previously in Figures 39a, 40a and 41a. In contrast, for SF12, and for the same N values, there is a higher collision ratio and ED drop ratio not only for high packet sizes, but also for low ones (see Figures 42b and 44b). The collision ratio is close to 1, due to the high frame transmission time for SF12 and the high number of EDs competing for transmission resources. The high collision and ED drop ratios, combined with a decreasing gateway drop ratio (see Figure 43b) lead to a relatively constant number of retransmissions (see Figure 45b), yielding a monotonically decreasing EPB as a function of packet size (see Figures 39b, 40b and 41b). For SF7 and N = 100, as illustrated in Figures 39a, 40a, and 41a, EPB decreases monotonically with packet size. This occurs because, in this scenario, the lower amount of EDs produces lower offered traffic load. While the collision ratio and the ED drop ratio increase with packet size, they remain low (compared to those obtained for greater N values, see Figures 42a and 44a). Note that, in consequence, the gateway drop ratio remains low and near-independent of packet size (see Figure 43a), whereas PDR is high and almost constant as well as a function of packet size (see Figures 46a and 47a). As a result, the number of ED retransmissions remains relatively low, it increases only slightly with packet size, and packet size itself dominates the decrease of EPB in this case (Figures 39a, 40a, and 41a). For SF12 and N = 100, the retransmission ratio remains constant with packet size, therefore EPB decreases monotonically with packet size (Figures 39a, 40a, and 41a). Note that, for SF7, the retransmission ratio is between 82.1% and 87.3% (see Figure 45a), for N ³ 1000 and for all packet sizes considered, which means that almost all transmissions are retransmissions and suggests that the system is in a saturation state. In the worst case, for each packet to be transmitted, there is one first attempt and 7 frame retransmissions, which leads to an upper bound for the retransmission ratio of 87.5 %. This is also the case for SF12 and all the considered N values, where the retransmission ratio is around 87.4% (see Figure 45b). Contributions to the Evaluation and Improvement of LoRaWAN 64 (a) (b) Figure 42. Collision ratio, for N = NC: (a) SF7, (b) SF12. (a) (b) Figure 43. Gateway drop ratio, for N = NC: (a) SF7, (b) SF12. 5. Energy efficiency-optimal packet size in LoRaWAN 65 (a) (b) Figure 44. ED drop ratio, for N = NC: (a) SF7, (b) SF12. 5.2.3. PDR in confirmed transmission mode We next discuss the obtained PDR results in conjunction with EPB performance. As shown in Figure 10a, for SF7 and TP = 1000 s, PDR decreases with packet size. This is mainly due to the increasing collision ratio and ED drop ratio (Figures 42a and 44a), and it is aggravated by gateway drops (Figure 43a). For SF7 and TP = 4000 s (see Figure 47a), PDR decreases more slowly with packet size, due to the lower offered network load. (a) (b) Figure 45. Retransmission ratio from EDs, for N = NC: (a) SF7, (b) SF12. Contributions to the Evaluation and Improvement of LoRaWAN 66 (a) (b) Figure 46. PDR and EPB for TP = 1000 s: (a) SF7, (b) SF12. For SF7 and N = 100, EPB decreases as well with packet size, showing a trade-off between PDR and EPB in the considered scenarios (Figures 10a and 11a). However, for N = 1000 and N = 2000, and for TP = 1000 s (Figure 10a), due to the “U” shape of EPB, the latter decreases with packet size only up to a packet size of ~80 bytes and ~40 bytes, respectively. These packet sizes represent threshold values that must not be exceeded for the sake of PDR and EPB, since further increasing packet size harms both performance parameters. Similar behavior occurs for TP = 4000 s and N = 2000 (Figure 47a), where the threshold packet size is ~100 bytes. Use of SF12 (Figure 10b and 47b) leads to significantly lower PDR than that achieved for SF7 (Figure 46a and 47a), especially for N ³ 1000, where PDR approaches zero. Except for N = 100, modifying the packet size does not vary the PDR significantly, although increasing packet size (up to 51 bytes for SF12) decreases EPB. 5.2.4. Mixed CED and UED scenarios In this subsection, we evaluate mixed scenarios with NC CEDs, and N – NC UEDs. Our aim is to study network performance in scenarios where at least a subset of the EDs (i.e., the UEDs) contribute less traffic load to the network, since they do not perform retransmissions. We analyze two main cases: NC = 100 and NC =1000, for different N values. For all of them, we also consider SF7 and SF12, and TP = 4000 s. Figures 48a and 49a show EPB for NC = 100 and NC = 1000, respectively, for SF7, and for different N values. For NC = 100 (Figure 48a), EPB decreases dramatically with packet size, even for high numbers of UEDs. In contrast, NC = 1000 (Figure 49a) shows a different behavior, with a slight EPB increase for high packet size and high N values, which produces an “asymmetric U” shape curve with a minimum for a medium packet size value. 5. Energy efficiency-optimal packet size in LoRaWAN 67 The EPB increase in Figure 49a, for high packet sizes, is due to a high collision ratio, even greater than 80% (see Figure 50a). The latter is due to the high number of EDs (from N = 1000 to N = 5000) and because a significant part of them (NC = 1000) use CTM. The greater number of collisions increases the amount of retransmissions performed by CEDs. This also causes a gateway drop ratio decrease (see Figure 51a) because a lower number of ACK frames need to be transmitted by the NS (due to a lower number of uplink frames reaching it). (a) (b) Figure 47. PDR and EPB for TP = 4000 s: (a) SF7, (b) SF12. (a) (b) Figure 48. EPB as a function of packet size, for TP = 4000 s, NC = 100 and several N values: (a) SF7, (b) SF12. Contributions to the Evaluation and Improvement of LoRaWAN 68 (a) (b) Figure 49. EPB as a function of packet size, for TP = 4000 s, NC = 1000 and several N values: (a) SF7, (b) SF12. (a) (b) Figure 50. Collision ratio in the gateway for TP = 4000 s, NC = 1000 and several N values: (a) SF7, (b) SF12. On the other hand, for SF7, packet drop ratio at EDs is low in all cases (below 10%, see Figure 52a) due to the relatively low load for each ED of a packet transmission every 4000 s, and the low transmission times due to SF7, in relation to duty cycle limitations. 5. Energy efficiency-optimal packet size in LoRaWAN 69 (a) (b) Figure 51. Gateway drop ratio for TP = 4000 s, NC = 1000 and several N values: (a) SF7, (b) SF12. (a) (b) Figure 52. ED drop ratio for TP = 4000 s, NC = 1000 and several N values: (a) SF7, (b) SF12. When SF12 is used (see Figures 48b and 49b), the EPB is greater than for SF7 as expected, due to the greater transmission time. For NC = 1000, EPB tends to decrease with packet size (Figure 49b). However, for NC = 100 and high N values (i.e., N = 2000), see Figure 48b, EPB follows an “asymmetric U” shape, with an optimal packet size (that minimizes EPB) of ~40 bytes. On the other hand, EPB decreases with N, from N = 120 to N = 500, while for greater N values (N ³ 1000), EPB increases with N (Figure 48b). We next explain the reasons for such behavior. For the lower N values considered, as N increases, the collision ratio does not vary significantly (Figure 53b). However, the ED drop ratio decreases with N (Figure 54b), due to the greater number of UEDs, which are less affected by the duty cycle Contributions to the Evaluation and Improvement of LoRaWAN 70 restrictions than CEDs. As a result, for low N values, EPB decreases with N. As N increases further (e.g., N ≥ 500), the collision ratio increases remarkably, which despite the ED drop ratio decrease with N, produces the observed EPB increase with N for N ≥ 1000. For NC = 1000, we can also highlight that the gateway drop ratio is similar to the one for SF7 (Figure 51), mainly due to a very high collision probability for SF12 (see Figure 50b), as a result of its greater transmission time compared with SF7. This reduces the amount of ACK frames to be transmitted to EDs from the NS, which are less affected by duty cycle restrictions at the gateway. For SF12, ED packet drops (see Figure 52b) show greater values than for SF7, due to the greater transmission time of the former. Finally, we focus on PDR and EPB in the mixed CED and UED scenarios. For SF7 and NC = 1000, since both PDR (Figure 55) and EPB (Figure 48a) decrease with packet size, a trade-off between PDR and EPB exists. CEDs outperform UEDs by 10%-20% (see Figure 55). For SF12 and NC = 1000, PDR is very low, below 5% in most cases, for both CEDs and UEDs, which illustrates that the offered load significantly exceeds the network capacity of the considered scenario in this case (Figure 56). For NC = 100 (Figures 57 and 58), the PDR is greater than that achieved for NC = 1000 (Figures 55 and 56) in the studied scenarios, due to the lower number of CEDs. For SF7 and NC = 100, CEDs show a PDR close to 100% for almost all packet sizes considered (Figure 57a). However, UEDs exhibit a PDR with a remarkable dependency on packet size for high N values (e.g., N > 500), which can fall to ~0.6 in the worst case (see Figure 57b). For SF12 and NC = 100, similar PDR is obtained by both CEDs and UEDs for low N values (e.g., N < 500), see Figure 58. For N ≥ 500, CEDs have a better PDR by 10%-25% (Figure 58). Since for SF12 and high N values EPB follows an “asymmetric U” shape (Figure 48b), a packet size greater than the EPB-optimal one (i.e., ~40 bytes for N = 2000) should not be exceeded, as PDR decreases with packet size. (a) (b) Figure 53. Collision ratio for TP = 4000 s, NC = 100 and several N values: (a) SF7, (b) SF12. 5. Energy efficiency-optimal packet size in LoRaWAN 71 (a) (b) Figure 54. ED drop ratio for TP = 4000 s, NC = 100 and several N values: (a) SF7, (b) SF12. (a) (b) Figure 55. PDR for TP = 4000, NC = 1000 and several N values, and SF7: (a) CEDs, (b) UEDs. Contributions to the Evaluation and Improvement of LoRaWAN 78 6.2.1. Extending the energy consumption empirical analysis to other LoRaWAN ED hardware platforms and network behaviors In the last years, the variety of available LoRaWAN ED hardware platforms has increased, with improved characteristics in terms of energy consumption and transmission efficiency. We propose to extend the energy consumption empirical analysis to other popular LoRaWAN ED platforms in the market, in order to characterize each model and produce an average model, intended to be more general and representative. On the other hand, there exist LoRaWAN networks which use parameter settings different from the traditionally typical ones (e.g., TTN networks currently use RECEIVE_DELAY1 and RECEIVE_DELAY2 of 5 and 6 seconds, respectively [90]). 6.2.2. Considering a greater set of transmission channels The number of transmission channels can be a key parameter in the performance of LoRaWAN in terms of energy modelling, SF12 Well behavior or in energy efficiency versus packet size. We propose to extend our study to 8 channels and beyond, as the majority of nowadays LoRaWAN EDs and gateways can work with, at least, 8 channels. 6.2.3. Analyzing the behavior of SF12 Well in a channel with losses In Chapter 4 we analyzed the SF12 Well phenomenon in a controlled channel with no errors or losses. In those conditions, the SF12 Well is related to a congestion situation. However, considering more realistic channel conditions, we propose to evaluate the influence of non-zero loss rate over the SF12 Well, and how to balance it with PDR. 6.2.4. Considering new strategies for improving EPB efficiency In Chapter 4 we considered some basic strategies intended to face the SF12 Well problem. The strategies were based on how SF is used and how the retransmission mechanism is applied. We propose to explore some alternatives that can balance complexity and energy efficiency improvement. The alternatives consist on the use of a limited number of SFs (for example, the 2 or 3 lowest SF values), applied to the three SFM proposed methods. We expect that the collision ratio will decrease enough to overcome the energy cost penalty due to ToA increase. 6.2.5. Improving PHY layer implementation in AFLoRa The study done in Chapters 4 and 5 is based on simulations using the AFLoRa framework, over the OMNeT++ platform simulator. The AFLoRa framework is built with a wireless module comprising a physical layer that applies a general behavior in terms of collision condition determination. We propose to improve the simulator with a more accurate physical layer implementation, taking into account the techniques used in real ED platforms in order to determine whether a packet has been received or not. Contributions The main contributions of the present PhD thesis can be summarized in the following publications. Published journal papers: 1. L. Casals, B. Mir, R. Vidal, and C. Gomez, “Modeling the Energy Performance of LoRaWAN,” Sensors, vol. 17, no. 10, p. 2364, Oct. 2017, doi: 10.3390/s17102364. Available online: http://dx.doi.org/10.3390/s17102364. 2. L. Casals, C. Gomez, and R. Vidal, “The SF12 Well in LoRaWAN: Problem and End-DeviceBased Solutions,” Sensors, vol. 21, no. 19, p. 6478, Sep. 2021, doi: 10.3390/s21196478. Available online: http://dx.doi.org/10.3390/s21196478. Submitted journal papers: 1. L. Casals, C. Gomez, and R. Vidal, “Understanding the Impact of Packet Size on the Energy Efficiency of LoRaWAN”, Journal of Communications and Networks, submitted: 2-3-2023. Other contributions from the present PhD thesis are the following improvements of FLoRa simulator (sorted by release date): 1. AFLoRa v0.9, Advanced Framework for LoRaWAN (v0.9). Available on Github: https://github.com/lluiscas/AFLoRa, 2021 (accessed on 30 December 2022). 2. AFLoRa v1.0, Advanced Framework for LoRaWAN v1.0. Available on Github: https://github.com/lluiscas/AFLoRa-v1, 2022 (accessed on 30 December 2022). References [1] Raza, U.; Kulkarni, P.; Sooriyabandara, M. Low Power Wide Area Networks: An Overview. IEEE Commun. Surv. Tutor. 2017, 19, 855–873, doi:10.1109/COMST.2017.2652320. [2] Farrell, S. Low-Power Wide Area Network (LPWAN) Overview. RFC8376, May 2018. [3] Mekki K.; Bajic E.; Chaxel F.; Meyer F. A comparative study of LPWAN technologies for large-scale IoT deployment. ICT Express, 2018. [4] Adelantado, F.; Vilajosana, X.; Tuset-Peiro, P.; Martinez, B.; Melia, J. Understanding the limits of LoRaWAN. Available online: https://arxiv.org/pdf/1607.08011 (accessed on 17 January 2023). [5] Augustin, A.; Yi, J.; Clausen, T. A study of LoRa: Long range & low power networks for the Internet of Things. Sensors 2016, 16, 1466, doi:10.3390/s16091466. [6] Sornin, N.; Yegin, A. LoRa Alliance Technical committee. LoRaWAN™ 1.1 Specification; LoRa Alliance Technical Committee: Beaverton, OR, USA, October, 2017. [7] LoRaWAN™ Certified Products. Available online: https://lora-alliance.org/showcase/search/?_sfm_lorawan_certified_device=certified (accessed on 17 January 2023). [8] Petajajarvi, J.; Mikhaylov, K.; Roivainen, A.; Hanninen, T.; Pettissalo, M. On the coverage of LPWANs: Range evaluation and channel attenuation model for LoRa technology. In Proceedings of the 14th International Conference on ITS Telecommunications (ITST), Copenhagen, Denmark, 2–4 December 2015; pp. 55–59. [9] Nolan, K.E.; Guibene, W.; Kelly, M.Y. An evaluation of low power wide area network technologies for the Internet of Things. In Proceedings of the IEEE International of Wireless Communications and Mobile Computing Conference (IWCMC), Paphos, Cyprus, 5–9 September 2016; pp. 439–444. [10] Petajajarvi, J.; Mikhaylov, K.; Pettissalo, M.; Janhunen, J.; Iinatti, J. Performance of a low-power wide-area network based on LoRa technology: Doppler robustness, scalability, and coverage. Int. J. Distrib. Sens. Netw. 2017, 13, 1–16. [11] Mikhaylov, K.; Petajajarvi, J.; Janhunen, J. On LoRaWAN Scalability: Empirical Evaluation of Susceptibility to Inter-Network Interference. Available online: https://arxiv.org/pdf/1704.04257 (accessed on 17 January 2023). [12] Haxhibeqiri, J.; Van den Abeele, F.; Moerman, I.; Hoebeke, J. LoRa Scalability: A Simulation Model Based on Interference Measurements. Sensors 2017, 17, 1193. [13] Petajajarvi, J.; Mikhaylov, K.; A.; Yasmin, R.; Hämäläinen, M.; Iinatti, J. Evaluation of LoRa LPWAN Technology for Indoor Remote Health and Wellbeing Monitoring. Int. J. Wirel. Inf. Netw. 2017, 24, 153–165. [14] Kim, B.; Hwang, K. Cooperative Downlink Listening for Low-Power Long-Range Wide-Area Network. Sustainability 2017, 9, 627, doi: 10.3390/su9040627. [15] 1.14., 3.14. Mahmoud, M.S.; Mohamad, A.A. H. A Study of Efficient Power Consumption Wireless Communication Techniques/ Modules for Internet of Things (IoT) Applications. Adv. Intern. Things 2016, 6, 19–29. [16] 1.15., 3.15. Bjelcevic, S.; Jemson, J.; Karusala, N.; Purcell, D. LAMBS: Light and Motion Based Safety. Available online: https://andyhub.com/wordpress/wp-content/uploads/LAMBSFinalReport.pdf (accessed on 17 January 2023). [17] Bor, M.; Roedig, U. LoRa Transmission Parameter Selection. In Proceedings of the 13th IEEE International Conference on Distributed Computing in Sensor Systems (DCOSS), Ottawa, ON, Canada, 5–7 June 2017. [18] Mendívil, L.J. Comparación de Soluciones Basadas en LPWAN e IEEE 802.15.4 Para Aplicaciones de Salud Móvil (“m-Health”). Master’s Thesis, Centro de Investigación Científica y de Educación Superior de Contributions to the Evaluation and Improvement of LoRaWAN 82 Ensenada, Ensenada, Mexico, 2017. Available online: https://cicese.repositorioinstitucional.mx/jspui/handle/1007/1274 (accessed on 17 January 2023). [19] Neumann, P.; Montavont, J.; Noël, T. Indoor deployment of low-power wide area networks (LPWAN): A LoRaWAN case study. In Proceedings of the IEEE 12th International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), New York, NY, USA, 17–19 December 2016; pp. 1–8. [20] Mikhaylov, K.; Petäjäjärvi, J. Design and implementation of the plug&play enabled flexible modular wireless sensor and actuator network platform. Asian J. Control, 2017, 19, 1392–1412. [21] Magno, M.; Aoudia, F.A.; Gautier, M.; Berder, O.; Benini, L. WULoRa: An energy efficient IoT end-node for energy harvesting and heterogeneous communication. In Proceedings of the IEEE Design, Automation & Test in Europe Conference & Exhibition, Lausanne, Switzerland, 27–31 March 2017; pp. 1528–1533. [22] Dongare, A.; Hesling, C.; Bhatia, K.; Balanuta, A.; Pereira, R.L.; Iannucci, B.; Rowe, A. OpenChirp: A LowPower Wide-Area Networking architecture. In Proceedings of the IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops), Kona, HI, USA, 13–17 March 2017; pp. 569–574. [23] Conus, G.; Lilis, G.; Zanjani, N.A.; Kayal, M. An event-driven low power electronics for loads metering and control in smart buildings. In Proceedings of the Second International Conference on IEEE Event-Based Control, Communication, and Signal Processing (EBCCSP), Krakow, Poland, 13–15 October 2016; pp. 1–7, doi: 10.1109/EBCCSP.2016.7605091. [24] Kim, C. AN079-Measuring Power Consumption of CC2530 with Z-Stack. Texas Instruments. Available online: http://www.ti.com/lit/an/swra292/swra292.pdf (accessed on 17 January 2023). [25] Aguilar, S.; Vidal, R.; Gomez, C. Opportunistic Sensor Data Collection with Bluetooth Low Energy. Sensors 2017, 17, 159, doi: 10.3390/s17010159. [26] Sartori, D.; Brunelli, D. A smart sensor for precision agriculture powered by microbial fuel cells. In Proceedings of the IEEE Sensors Applications Symposium (SAS), Catania, Italy, 20–22 May 2016; pp. 1–6, doi: 10.1109/SAS.2016.7479815. [27] Toussaint, J.; El Rachkidy, N.; Guitton, A. Performance analysis of the on-the-air activation in LoRaWAN. In Proceedings of the IEEE 7th Annual Information Technology, Electronics and Mobile Communication Conference (IEMCON), Vancouver, BC, Canada, 13–15 November 2016; pp. 1–7, doi: 10.1109/IEMCON.2016.7746082. [28] Slabicki, M.; Premsankar, G.; Di Francesco, M. Adaptive Configuration of Lora Networks for Dense IoT Deployments. In Proceeding of the NOMS 2018-2018 IEEE/IFIP Network Operations and Management Symposium, Taipei, Taiwan, 23–27 April 2018; pp 1–9, doi:10.1109/NOMS.2018.8406255. [29] Casals, L.; Mir, B.; Vidal, R.; Gomez, C. Modeling the Energy Performance of LoRaWAN. Sensors 2017, 17, 2364, doi:10.3390/s17102364. [30] Yegin, A.; Seller, O.; LoRaWAN® L2 1.0.4 Specification (TS001-1.0.4), LoRa Alliance, October 2020. [31] Semtech Corporation. SX1272/3/6/7/8: LoRa Modem. Designer’s Guide. AN1200.13; Semtech Corporation: Camarillo, CA, USA, 2013. [32] Alliance Technical committee Regional Parameters Workgroup. LoRaWAN™ 1.1 Regional Parameters; LoRa Alliance Technical Committee: Beaverton, OR, USA, 2018. [33] LoRa Alliance Technical Committee. LoRaWAN™ 1.0.2 Regional Parameters; LoRa Alliance Technical Committee: Fremont, CA, USA, 2020. [34] ETSI EN 300.220-2 v3.2.1 (2018-06). Short Range Devices (SRD) Operating in the Frequency Range 25 MHz to 1000 MHz; Part 2: Harmonised Standard for Access to Radio Spectrum for Non Specific Radio Equipment. Available online: https://www.etsi.org/deliver/etsi_en/300200_300299/30022002/03.02.01_60/en_30022002v030201p.pdf (accessed on 17 January 2023). [35] Microchip RN2483. Available online: https://ww1.microchip.com/downloads/aemDocuments/documents/OTH/ProductDocuments/DataSheets/RN2 483-Low-Power-Long-Range-LoRa-Technology-Transceiver-Module-DS50002346F.pdf (accessed on 17 January 2023). [36] Errata Microchip RN2483. Available online: http://ww1.microchip.com/downloads/en/devicedoc/80000689a.pdf (accessed on 17 January 2023). [37] LoRa Alliance Technical committee. LoRaWAN Specification; Version V1.0.3; LoRa Alliance: Beaverton, OR, USA, 2018. References 83 [38] LoRa Alliance Certification. Available online: https://lora-alliance.org/lorawan-certification (accessed on 17 January 2023). [39] Sornin, N.; Luis, M.; Eirich, T.; Kramp, T.; Hersent, O. LoRaWAN Specification; Version V1.0.2; LoRa Alliance: Beaverton, OR, USA, 2016. [40] LoRa Alliance Technical committee. LoRaWAN™ 1.0.2 Regional Parameters; LoRa Alliance Technical Committee: Beaverton, OR, USA, 2017. [41] Mikhaylov, K.; Petajajarvi, J.; Hanninen, T. Analysis of capacity and scalability of the LoRa low power wide area network technology. In Proceedings of the 22th European Wireless Conference, Oulu, Finland, 18–20 May 2016; pp. 1–6. [42] LoRaWAN Multitech mDot. Available online: http://www.multitech.com/documents/publications/datasheets/86002171.pdf (accessed on 17 January 2023). [43] NetBlocks XRange. Available online: https://www.netblocks.eu/xrange-sx1272-lora-datasheet/ (accessed on 17 January 2023). [44] Microchip RN2483 LoRa Mote. Available online: https://www.microchip.com/en-us/development-tool/dm164138 (accessed on 17 January 2023). [45] LoRaBug. Available online: https://github.com/OpenChirp/LoRaBug (accessed on 17 January 2023). [46] Semtech SX1272. Datasheet: SX1272_DS_V4.pdf. Available online: https://www.semtech.com/products/wireless-rf/lora-connect/sx1272#documentation (accessed on 17 January 2023). [47] Semtech SX1276. Datasheet: DS_SX1276-7-8-9_W_APP_V7.pdf. Available online: https://www.semtech.com/products/wireless-rf/lora-connect/sx1276#documentation (accessed on 17 January 2023). [48] HopeRF HM-TRLR-LF/HFS Series 100 mW. Available online: https://www.hoperf.com/data/upload/portal/20190307/HM-TRLR-S_Series_EN.pdf (accessed on 17 January 2023). [49] Kerlink LoRa IoT Station. Available online: http://www.kerlink.com/product/wirnet-station/ (accessed on 17 January 2023). [50] Seller, O.; Sornin, N. Low Power Long Range Transmitter. U.S. Patent 9,252,834, 6 August 2014. [51] Haxhibeqiri, J.; De Poorter, E.; Moerman, I.; Hoebeke, J. A survey of LoRaWAN for IoT: From technology to application. Sensors 2018, 18, 3995, doi:10.3390/s18113995. [52] Pop, A.; Raza, U.; Kulkarni, P.; Sooriyabandara, M. Does Bidirectional Traffic Do More Harm Than Good in LoRaWAN Based LPWA Networks? GLOBECOM 2017 - 2017 IEEE Global Communications Conference, 2017, pp. 1–6, doi:10.1109/GLOCOM.2017.8254509. [53] Capuzzo, M.; Magrin, D.; Zanella, A. Confirmed Traffic in LoRaWAN: Pitfalls and Countermeasures. In Proceedings of the 2018 17th Annual Mediterranean Ad Hoc Networking Workshop (Med-Hoc-Net), Capri, Italy, 20-22 June 2018; pp. 1–7, doi:10.23919/MedHocNet.2018.8407095. [54] Magrin, D.; Capuzzo, M.; Zanella, A. A Thorough Study of LoRaWAN Performance Under Different Parameter Settings. IEEE Internet Things J. 2020, 7, 116–127, doi:10.1109/JIOT.2019.2946487. [55] Lavric, A.; Popa, V. Performance Evaluation of LoRaWAN Communication Scalability in Large-Scale Wireless Sensor Networks. Wirel. Commun. Mob. Comput. 2018, 2018, doi:10.1155/2018/6730719. [56] Böcker, S.; Arendt, C.; Jörke, P.; Wietfeld, C. LPWAN in the Context of 5G: Capability of LoRaWAN to Contribute to MMTC. In Proceedings of the 2019 IEEE 5th World Forum on Internet of Things (WF-IoT), Limerick, Ireland, 15–18 April 2019; pp. 737–742, doi:10.1109/WF-IoT.2019.8767333. [57] FLoRa. Framework for LoRa. Available online: https://flora.aalto.fi (accessed on 17 January 2023). [58] AFLoRa. Advanced Framework for LoRaWAN. https://github.com/lluiscas/AFLoRa (accessed on 17 January 2023). [59] Capuzzo, M.; Magrin, D.; Zanella, A. Mathematical Modeling of LoRa WAN Performance with BiDirectional Traffic. In Proceedings of the 2018 IEEE Global Communications Conference (GLOBECOM), Abu Dhabi, UAE, 9–13 December 2018; pp. 206–212, doi:10.1109/GLOCOM.2018.8647351. [60] Mikhaylov, K.; Petäjäjärvi, J.; Pouttu, A. Effect of Downlink Traffic on Performance of LoRaWAN LPWA Networks: Empirical Study. In Proceedings of the 2018 IEEE 29th Annual International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Bologna, Italy, 9–12 September 2018; pp. 1– 6, doi:10.1109/PIMRC.2018.8580721. [61] Di Vincenzo, V.; Heusse, M.; Tourancheau, B. Improving Downlink Scalability in LoRaWAN. In Proceedings of the ICC 2019-2019 IEEE International Conference on Communications (ICC), Shanghai, China, 20–24 May 2019; pp. 1–7, doi:10.1109/ICC.2019.8761157. Contributions to the Evaluation and Improvement of LoRaWAN 84 [62] Sanchez-Iborra, R.; Sanchez-Gomez, J.; Ballesta-Viñas, J.; Cano, M.-D.; Skarmeta, A.F. Performance Evaluation of LoRa Considering Scenario Conditions. Sensors 2018, 18, 772, doi:10.3390/s18030772. [63] Hasegawa, Y.; Suzuki, K. A Multi-User ACK-Aggregation Method for Large-Scale Reliable LoRaWAN Service. In Proceedings of the ICC 2019-2019 IEEE International Conference on Communications (ICC), Shanghai, China, 20–24 May 2019; pp. 1–7, doi:10.1109/ICC.2019.8761253. [64] LoRa Alliance Technical Committee. LoRaWAN Specification; Version V1.1; LoRa Alliance: Beaverton, OR, USA, 2017. [65] Kufakunesu, R.; Hancke, G.P.; Abu-Mahfouz, A.M. A Survey on Adaptive Data Rate Optimization in LoRaWAN: Recent Solutions and Major Challenges. Sensors 2020, 20, 5044, doi:10.3390/s20185044. [66] Garlisi, D.; Tinnirello, I.; Bianchi, G.; Cuomo, F. Capture Aware Sequential Waterfilling for LoRaWAN Adaptive Data Rate. IEEE Trans. Wirel. Commun. 2021, 20, 2019–2033, doi:10.1109/TWC.2020.3038638. [67] Reynders, B.; Meert, W.; Pollin, S. Power and Spreading Factor Control in Low Power Wide Area Networks. In Proceedings of the 2017 IEEE International Conference on Communications (ICC), Paris, France, 21–25 May 2017; pp. 1–6. [68] Sallum, E.; Pereira, N.; Alves, M.; Santos, M. Improving Quality-Of-Service in LoRa Low-Power Wide-Area Networks through Optimized Radio Resource Management. J. Sens. Actuator Netw. 2020, 9, 10, doi:10.3390/jsan9010010. [69] Bankov, D.; Khorov, E.; Lyakhov, A. LoRaWAN Modeling and MCS Allocation to Satisfy Heterogeneous QoS Requirements. Sensors 2019, 19, 4204, doi:10.3390/s19194204. [70] Kim, D.-Y.; Kim, S.; Hassan, H.; Park, J.H. Adaptive Data Rate Control in Low Power Wide Area Networks for Long Range IoT Services. J. Comput. Sci. 2017, 22, 171–178, doi:10.1016/j.jocs.2017.04.014. [71] Cattani, M.; Boano, C.A.; Römer, K. An Experimental Evaluation of the Reliability of LoRa Long-Range Low-Power Wireless Communication. J. Sens. Actuator Netw. 2017, 6, 7, doi:10.3390/jsan6020007. [72] Mikhaylov, K.; Stusek, M.; Masek, P.; Fujdiak, R.; Mozny, R.; Andreev, S.; Hosek, J. On the Performance of Multi-Gateway LoRaWAN Deployments: An Experimental Study. In Proceedings of the 2020 IEEE Wireless Communications and Networking Conference (WCNC), Seoul, Korea, 25–28 May 2020; pp. 1–6. [73] Li, L.; Ren, J.; Zhu, Q. On the Application of LoRa LPWAN Technology in Sailing Monitoring System. In Proceedings of the 2017 13th Annual Conference on Wireless On-demand Network Systems and Services (WONS), Jackson Hole, WY, USA, 21–24 February 2017; pp. 77–80. [74] Di Renzone, G.; Parrino, S.; Peruzzi, G.; Pozzebon, A.; Bertoni, D. LoRaWAN Underground to Aboveground Data Transmission Performances for Different Soil Compositions. IEEE Trans. Instrum. Meas. 2021, 70, 1– 13, doi:10.1109/TIM.2021.3061820. [75] Alves, H.B.M.; Lima, V.S.S.; Silva, D.R.C.; Nogueira, M.B.; Rodrigues, M.C.; Cunha, R.N.; Carvalho, D.F.; Sisinni, E.; Ferrari, P. Introducing a Survey Methodology for Assessing LoRaWAN Coverage in Smart Campus Scenarios. In Proceedings of the 2020 IEEE International Workshop on Metrology for Industry 4.0 IoT, Roma, Italy, 3–5 June 2020; pp. 708–712. [76] Kim, B. and Hwang, K. Cooperative Downlink Listening for Low-Power Long-Range Wide-Area Network, Sustainability, vol. 9, no. 4, Art. no. 4, Apr. 2017, doi: 10.3390/su9040627. [77] Finnegan, J.; Brown. S. and Farrell, R. Modeling the Energy Consumption of LoRaWAN in ns-3 Based on Real World Measurements, 2018 Global Information Infrastructure and Networking Symposium (GIIS), 2018, pp. 1-4, doi: 10.1109/GIIS.2018.8635786. [78] Bouguera, T.; Diouris, J.-F.; Chaillout, J.-J. and Andrieux, G. Energy consumption modeling for communicating sensors using LoRa technology, 2018 IEEE Conference on Antenna Measurements & Applications (CAMA), 2018, pp. 1-4, doi: 10.1109/CAMA.2018.8530593. [79] Eriksen, R. Energy Consumption of Low Power Wide Area Network Node Devices in the Industrial, Scientific and Medical Band, Dissertation, 2019. [80] Pukrongta, N. and Kumkhet, B. The relation of LoRaWAN efficiency with energy consumption of sensor node, in 2019 International Conference on Power, Energy and Innovations (ICPEI), Pattaya, Chonburi, Thailand, Oct. 2019, pp. 90–93, doi: 10.1109/ICPEI47862.2019.8945016. [81] Singh, R. K.; Puluckul, P. P.; Berkvens, R. and Weyn, M. Energy Consumption Analysis of LPWAN Technologies and Lifetime Estimation for IoT Application, Sensors, vol. 20, no. 17, p. 4794, Aug. 2020, doi: 10.3390/s20174794. [82] Sherazi, H. H. R.; Grieco, L. A.; Imran, M. A. and Boggia, G. Energy-Efficient LoRaWAN for Industry 4.0 Applications, in IEEE Transactions on Industrial Informatics, vol. 17, no. 2, pp. 891-902, Feb. 2021, doi: 10.1109/TII.2020.2984549. References 85 [83] Nurgaliyev, M. K.; Saymbetov, A. K.; Zholamanov, B. N.; Yeralkhanova, A. Т. and Zhuman, G. B. Predicting the Lifetime of Lora Based Wireless Sensor Networks Using a Probabilistic Model of Markov Chains, News of NAN RK. SGTS, vol. 2, no. 336, pp. 157–164, Apr. 2021, doi: 10.32014/2021.25181726.36. [84] Marini, R.; Mikhaylov, K.; Pasolini, G. and Buratti, C. LoRaWANSim: A Flexible Simulator for LoRaWAN Networks, Sensors, vol. 21, no. 3, p. 695, Jan. 2021, doi: 10.3390/s21030695. [85] Maudet, S.; Andrieux, G.; Chevillon, R. and Diouris, J.-F. Refined Node Energy Consumption Modeling in a LoRaWAN Network, Sensors, vol. 21, no. 19, p. 6398, Sep. 2021, doi: 10.3390/s21196398. [86] Taleb, H.; Nasser, A.; Andrieux, G.; Charara, N. and Cruz, E. M. Energy Consumption Improvement of a Healthcare Monitoring System: Application to LoRaWAN, IEEE Sensors J., vol. 22, no. 7, pp. 7288–7299, Apr. 2022, doi: 10.1109/JSEN.2022.3150716. [87] Casals, L.; Gomez, C.; Vidal, R. The SF12 Well in LoRaWAN: Problem and End-Device-Based Solutions, Sensors, vol. 21, no. 19, p. 6478, Sep. 2021, doi: 10.3390/s21196478. [88] AFLoRa v1.0, available on Github: https://github.com/lluiscas/AFLoRa-v1 (accessed on 17 January 2023) [89] LoRaWAN® Regional Parameters RP002-1.0.4, LoRa Alliance, 2022 [90] MAC Settings. The Things Stack v3.23.1. Available online: https://www.thethingsindustries.com/docs/devices/mac-settings/, (accessed on 17 January 2023)